EMPLOYEE ATTRITION PREDICTION USING MACHINE LEARNING MODELS: A REVIEW PAPER

International Journal of Artificial Intelligence and Applications (IJAIA), Vol.15, No.2, March 2024
DOI:10.5121/ijaia.2024.15202 23
EMPLOYEE ATTRITION PREDICTION USING
MACHINE LEARNING MODELS: A REVIEW PAPER
Haya Alqahtani, Hana Almagrabi and Amal Alharbi
Department of Information Systems, College of Computers and Information Technology,
King Abdulaziz University, Jeddah, Kingdom of Saudi Arabia
ABSTRACT
Employee attrition refers to the decrease in staff numbers within an organization due to various reasons. As
it has a negative impact on long-term growth objectives and workplace productivity, firms have recognized
it as a significant concern. To address this issue, organizations are increasingly turning to machine-learning
approaches to forecast employee attrition rates. This topic has gained significant attention from researchers,
especially in recent times. Several studies have applied various machine-learning methods to predict
employee attrition, producing different results depending on the employed methods, factors, and datasets.
However, there has been no comprehensive comparative review of multiple studies applying machine-
learning models to predict employee attrition to date. Therefore, this study aims to fill this gap by providing
an overview of research conducted on applying machine learning to predict employee attrition from 2019 to
February 2024. A literature review of relevant studies was conducted, summarized, and classified. Most
studies agree on conducting comparative experiments with multiple predictive models to determine the most
effective one. From this literature survey, the RF algorithm and XGB ensemble method are repeatedly the
best-performing, outperforming many other algorithms. Additionally, the application of deep learning to
employee attrition prediction issues also shows promise. While there are discrepancies in the datasets used
in previous studies, it is notable that the dataset provided by IBM is the most widely utilized. This study
serves as a concise review for new researchers, facilitating their understanding of the primary techniques
employed in predicting employee attrition and highlighting recent research trends in this field. Furthermore,
it provides organizations with insight into the prominent factors affecting employee attrition, as identified by
studies, enabling them to implement solutions aimed at reducing attrition rates.
KEYWORDS
Employee Attrition, Prediction, Machine Learning, Ensemble Methods, Feature Selection
1. INTRODUCTION
The workforce is the most valuable asset in any company or organization [1] [2]. It plays a
fundamental role in achieving the organization’s vision and mission [3]. Executives spend
significant time selecting qualified employees and investing in employee training [2]. However,
one of the most prominent issues organizations face is employee attrition [4].
Employee attrition is the term that refers to a decrease in the number of employees in a company
due to employees' voluntary or involuntary departures from their jobs [4] [5]. Employees choose
to leave their positions for various reasons, such as work pressure, an unsatisfactory work
environment, or inadequate compensation [6]. According to a McKinsey survey, nearly half (48%)
of workers are leaving their current roles to transition into a new industry [7].
The issue of employee attrition has garnered attention from many organizations due to its negative
impact [8]. Employee turnover rates have reached their highest levels in the past decade, forcing

24
many organizations to deal with this challenge [9]. This urgency prompts leaders to proactively
address turnover rates to mitigate their adverse effects on business activities and operations [9].
The impact of this issue can be evaluated using various metrics, such as sales growth, return on
equity, customer service quality, and profitability [4]. It has been shown that a high attrition rate
presents significant challenges for businesses, as companies invest significant resources—
financial, temporal, and material—in training employees for specific roles, incurring substantial
losses when these employees leave [10]. On the other hand, organizations are forced to allocate
additional resources to human capital, recruitment, orientation, and the development of new hires
[8] [11]. Importantly, the most significant expenses for organizations are investments in their
human resources, including employee benefits, training, salaries, bonuses, and other forms of
compensation [12].
Research indicates that voluntary employee turnover can cost an organization more than one and a
half times an employee’s annual salary, taking into account total expenses such as reassignment,
recruitment, and training of replacements [13]. Therefore, the significant impact of staff attrition
on firms highlights the increased interest in this topic [4].
In such a context, the importance of using analytics in the human resources sector becomes
increasingly clear, as it enables HR professionals to gain comprehensive insights into their
personnel and make informed decisions [11]. Employee analytics provides human resources
managers with easily accessible and understandable data about their workforce’s characteristics,
performance, and behavior, enhancing their understanding of their departments and employees
[14]. Descriptive analytics are used to analyze data or translate it into relevant information,
shedding light on past occurrences but offering limited insight into future events [14]. On the other
hand, predictive analytics have emerged as a valuable tool for forecasting future events, facilitating
improved and unbiased decision-making within organizations [14] [15]. By leveraging historical
data trends, machine-learning classifiers, such as data mining algorithms, help predict future
employee behaviors, such as performance and attrition rates. As a result, through the mining and
analysis of historical employee data, HR can anticipate future organizational behaviors [16].
The use of machine learning for predicting employee attrition has attracted considerable research
attention, especially in recent years. The growing amount of data in this field has resulted in
numerous studies investigating this topic [6].
Previous studies have used different machine-learning methods to forecast employee attrition, each
helping to improve solutions for this issue. However, the results of these studies have differed
based on the tools and methodologies used. Despite the substantial amount of research, as far as
we know, there is no comprehensive review that compares the different research efforts that have
employed machine-learning models to predict employee attrition.
Therefore, this study aims to provide an overview of the research conducted in the field of using
machine learning to predict employee attrition in recent years, from 2019 to February 2024. A
thorough literature review was conducted, gathering relevant studies published in electronic
journals, which were then summarized and categorized. The goal of this study is to offer a concise
review for new researchers, helping them understand the main approaches used in this field.
The rest of the paper is organized as follows. The second section discusses the issue of employee
attrition. The third section reviews the factors that influence employee attrition, as explored in
previous studies. The fourth section focuses on predicting employee attrition, including a
classification of previous work in this field, the approaches used, and a review of the resources and

25
techniques employed. The fifth section provides a comprehensive discussion of the previous
sections, while the paper concludes with Section 6.
2. EMPLOYEE ATTRITION AS A PROBLEM
Employee attrition refers to the decline in the workforce of an organization whether employees
leave the organization voluntarily or are retired. Employee turnover is the number of current
employees who are replaced by new hires for a set amount of time. High attrition leads to high
employee turnover in any organization [17]. "Turnover" and "attrition" are two trading terms that
are constantly at odds with one another. The term "attrition" describes a decline in the workforce.
While "Turnover" occurs when a company searches for a replacement for a departing employee.
These terms can be used interchangeably for workforce data analysis and other metrics required
for workforce planning [18].
There are two types of employee attrition: voluntary attrition and involuntary attrition. Workers
dismissed by their employers for a variety of reasons, such as poor performance or increased
workload, are said to have undergone involuntary attrition. Voluntary attrition is when high-
performing workers who choose to quit the company voluntarily do so despite the company's
efforts to keep them [19].
The main objectives of human resource management are to attract, inspire, and keep talent within
the company by creating management processes that guarantee that human talent is utilized
efficiently [20]. Although companies invest in their workforce by offering exceptional training and
a pleasant working environment because they understand the value of their employees, they also
experience voluntary attrition and the loss of brilliant individuals. Finding replacements is another
issue that comes with substantial expenditures for the business, such as those connected to
employment, training, and recruiting [19].
Employee attrition is one of the main issues facing businesses. According to statistics, 68 % of
workers quit their employment voluntarily because of a variety of problems that can be controlled
at work, such as poor management, a lack of benefits, and a lack of recognition [21]. High attrition
rates prove to be costly for the company as the company invests time, money, and assets to train
employees to make them job-ready in a particular company. If workers depart, the company not
only loses their valuable employees, but also the money invested in finding, choosing, and
preparing them for their positions. On the other side, the company must continually increase its
investment in new hire recruitment, training, and development [10] [22]. In addition, new hires will
also go through a similar learning curve as seasoned internal staff members to reach equivalent
proficiency levels in technical or business expertise [23]. In addition, this issue may cause
significant problems for organizations, such as lower earnings brought on by the departure of
productive workers [21].
When highly effective individuals leave the company in search of better opportunities, the losses
incurred when a competent employee quits are not limited to advanced product beliefs, great
project management, or connections with customers. This can hurt companies as their productivity
drops significantly, hampering organizational morale [10].
Additionally, employee attrition results in productivity and performance that are below optimal
levels [3], and loss of effective project management, which leads to the dissatisfaction of customers
and other stakeholders [4] [10]. It also risks leaking basic knowledge and technologies to
competitors [15]. Studies have also revealed that organizations with higher voluntary turnover rates
perform worse than their competitors, endangering their future market opportunities [13]. More
significantly, a company is more likely to experience reputational harm and harm to its brand, from

26
which it will be challenging to regain [9] [21]. Consequently, it makes sense that the reason behind
the great interest in employee attrition is that it has serious consequences for organizations [4].
It has become more important than ever for companies to do their best to reduce employee turnover,
given the proven detrimental effects of such turnover on their operations, and because of these
effects, by implementing regulations and creating policies that are more satisfactory to employees
and thus retain them. Companies can take advantage of the large amounts of employee data they
have to help address this problem.
3. LITERATURE REVIEW: FACTORS INFLUENCING EMPLOYEE ATTRITION
While human resources are indeed valuable assets crucial to a company’s success, one of the most
significant challenges in human resource management is employee attrition [24]. Addressing this
issue primarily involves identifying the underlying causes and researching associated factors. In
this section, we review studies that have identified or examined some of these factors and their
impact on employee attrition.
For instance, the authors in [25] identified five primary causes of voluntary employee turnover.
These include lack of recognition, insufficiently competitive compensation systems, recruitment
practices, managerial style, and a toxic work environment.
According to the study in [26], employee attrition occurs due to various reasons, including a lack
of support for employees in lower positions within the organization, inadequate communication
channels between these workers and their superiors, monotonous tasks that result in a loss of
creativity, limited career prospects, and a lack of opportunities for advancement.
The authors in [27] emphasized two types of factors that contribute to employee turnover. These
include job-related factors such as job stress, discordance regarding job tasks and duties, and job
dissatisfaction, as well as organizational factors such as organizational instability, deficient
communication systems, and limited employee involvement in decision-making processes.
Additionally, the authors in [28] confirmed that employees leave their positions due to feelings of
being underpaid, disagreements with management, issues with working hours, a lack of career
growth opportunities, or family-related concerns. Furthermore, in addition to the mismatch
between job roles and employee skills, other contributing factors include inadequate training,
limited opportunities for professional growth, lack of recognition or appreciation, poor work-life
balance resulting in excessive work-related stress, and diminished confidence in leadership [22].
Another study [29], identified administrative factors such as unstable management, inadequate
work environments, lack of essential facilities, low salaries, and insufficient financial incentives as
significant contributors to turnover. Additionally, the study highlighted further influences on the
turnover rate, including limited opportunities for career advancement, influence from coworkers,
alignment between the candidate and the job, and the availability of alternative job prospects.
According to the authors in [30], insufficient training and inadequate training duration can leave
workers feeling unprepared to meet the requirements of their roles. The lack of opportunities for
development and advancement, where organizations fail to provide career paths and training
opportunities for skilled and ambitious professionals, leads to the loss of talented employees.
Incorrect job profiles also contribute to high turnover rates. Additionally, ineffective management
and limited decision-making opportunities contribute to attrition [30].
Similarly, the author in [31] identified several causes of attrition, including a perceived lack of
career growth prospects, ineffective appraisal systems that fail to adequately recognize individual

27
efforts and contributions, management’s indifferent attitude towards employees, limited long-term
growth opportunities, inadequate provisions for ongoing learning, poor quality of HR practices,
ineffective leadership, uninspiring leadership leading to reduced enthusiasm at work, inflexible
work arrangements and inadequate compensation, excessive pressure with insufficient recognition,
failure of the organization to adhere to ethical practices and guidelines, and unsafe work
environments.
Another study [32] which focused on employees in private financial institutions, identified several
factors that contribute to attrition. The study concluded that the strained relationship between
management and employees, employee dissatisfaction, lack of job security, and job pressure were
significant contributors to employee attrition [32].
Furthermore, research conducted by several scholars has highlighted various factors that influence
turnover rates. These factors can be categorized into organizational factors, job-related factors,
work environment factors, and personal factors [33].
4. EMPLOYEE ATTRITION PREDICTION
Several factors, such as personal, job-related, and organizational characteristics, have been proven
to influence employee attrition, making them valuable predictors of attrition rates within an
organization [16].
The process of machine-learning prediction involves training an algorithm using a historical
dataset. Afterwards, for each record in new data, the algorithm generates potential values for
unknown variables [34].
managers and HR departments to assess an employee’s intention to leave the organization,
considering factors such as length of employment and performance. As a result, prediction models
can be created using historical data to evaluate employee performance, engagement, and behavior
within the organization [16].
Machine-learning models are trained on preprocessed data, allowing them to make informed
decisions when presented with new data. The primary objective of these models is to identify
patterns within the data and apply the knowledge gained to make predictions [1]. There are two
main types of models: supervised learning and unsupervised learning. In supervised learning, the
training set is labeled by humans, which enables the model to classify new data and predict
outcomes once it learns the relationship between input and output data. On the other hand,
unsupervised learning processes large amounts of data without human supervision, as there are no
labeled instances [35].
To validate machine-learning classification models developed using current employee data,
accuracy tests, and validations can be performed. This may involve using a different dataset or
testing the trained prediction model using 20% of the main dataset that was not used during the
training [36].
Efficiency and scalability are vital for machine-learning algorithms to effectively extract
information from large datasets or dynamic data streams. Metrics such as classification accuracy
and training time are crucial for evaluating the performance of data mining algorithms and selecting
the best algorithms for classification or prediction tasks [37].
Applying machine learning and data mining with various optimization algorithms has consistently
yielded valuable insights from different datasets [38]. Predictive analysis uses these algorithms to

28
forecast future trends, helping identify potential employees who may leave the company. This
proactive approach enables quick decision-making and reduces the loss of human resources [38]
[39], promoting continuous growth and maintaining a higher business [16].
The following subsections will review the types of datasets used in previous studies and the
approaches employed to predict employee attrition. Additionally, the focus will be on studies that
utilized ensemble methods to enhance forecasting performance and discussions on feature selection
techniques that improved prediction accuracy.
4.1.Datasets
Researchers have applied different machine-learning algorithms to different datasets in studies
predicting employee attrition. Some studies relied on previous ready-made and available databases
that were collected either for the study itself or other purposes, and some studies collected the study
database from real employee data. In this sub-section, previous work is classified according to the
database used:
4.1.1.Available Datasets
IBM HR Analytics Employee Attrition Dataset: The fictional dataset was created by IBM data
scientists in 2016 [21]. It consists of 1470 observations and includes 34 features that are relevant
to employees’ work and personal data. The dataset also includes a binary target variable called
“attrition,” where “yes” indicates an employee who left the company and “no” indicates an
employee who stayed [1]. However, the dataset is highly unbalanced, with 237 positive samples
(former employees) and 1233 negative samples (current employees), which significantly skews the
data [6]. Various researchers have conducted experiments using this dataset to predict employee
attrition, and these studies are referenced in [15], [22], [23], [40], and others.
Datasets Available on Kaggle: Some researchers have used pre-existing datasets available online,
such as those provided by the Kaggle website. For instance, the dataset utilized in [11] for human
resource management is easily accessible on kaggle.com and comprises approximately 14,000
records with ten attributes pertaining to employee attrition problems.
In [14], the performance of predictive models was assessed using two simulated human resources
datasets. The first dataset, obtained from Kaggle, consists of 15,000 samples with the target
variable “left” and nine features, including time spent in the company, work accidents, promotions
in the last five years, sales, satisfaction level, number of projects, average monthly hours, and
salary. The second dataset is a medium-sized dataset obtained from IBM.
Dataset from SAS: In [8] ], researchers utilized a real dataset from SAS, which consisted of 1.5k
rows and 35 variables/columns. The dataset included categorical and interval variables and was
sourced from the SAS library (www.sas.com).
4.1.2.Collected Employees Dataset
Many previous studies have collected datasets to analyze employee data and predict attrition. These
datasets were either sourced from specific institutions and organizations or gathered through
surveys that covered factors that could potentially contribute to workers leaving their jobs.
In [10], the authors proposed a method for predicting the rate of employee attrition and assessing
employees’ emotional well-being within the company. They collected data through a questionnaire
that included attrition-related questions and the Goldberg Depression Questionnaire. The dataset

29
considered various elements such as workplace equality, job satisfaction, overtime, working hours,
travel, personal life, and others. Additionally, the authors analyzed the depression questionnaire
using Goldberg’s Depression Questionnaire [10].
In [41], the authors describe and examine a Europe-wide study on employee turnover intention.
This study was based on the Global Employee Engagement Index (GEEI) survey conducted by
Effectory in 2018. The survey included 18,322 employees from 56 European countries and
consisted of 123 questions that covered socio-demographic, labor, and industry aspects, as well as
an objective question [41].
The study conducted by [4]. identified the most significant factors contributing to employee
attrition using data from the human resources department of an Iranian pharmaceutical company.
The dataset included 577 employee observations and comprised 31 characteristics, including nine
numeric-type variables, three logical variables, and 19 character-type variables [4].
In a separate study [42], the researchers investigated the correlation between employee
performance and organizational growth. For this research, data on employees was collected from
a well-established MNC business in Chennai. The company employs around 5000 individuals in
different departments, age groups, and locations [42].
In an Indian study [16], employee information was obtained from the HR datasets of three IT
companies in India. The dataset included 22 input features, and the employees’ job status at the
time of data capture was considered as the response variable [16].
In the study [14], an online questionnaire was created and used to collect information from
participants. The goal was to gather authentic employee data and identify factors contributing to
attrition. The collected features were divided into three sections using an exploratory approach: the
first section covered demographic variables, the second addressed overall levels of satisfaction,
motivation, engagement, and interest in life, and the third aimed to identify the elements
participants considered most significant in causing turnover. The database consisted of 450
responses from individuals across various universities worldwide, including Tunisia, Norway,
France, and the United States. Around 53% of participants expressed no intention to leave their
current positions, while 47% indicated a desire to do so [14].
To predict employee attrition, researchers in the study [43] used an authentic HR dataset from
employees of a Kazakhstani higher education institution. The dataset contained actual 2023 data
on employee demographics, job characteristics, gender, and other relevant factors. The dataset
included approximately 1500 observations, with instances of employee attrition accounting for
39% of the total [43].
4.2. Approaches
During the literature review of studies on predicting employee attrition, it is noteworthy that most
studies relied on either traditional machine-learning approaches or deep-learning methods. In this
subsection, we categorize previous studies based on the approach used to predict employee
attrition.
4.2.1.Machine Learning
In a study by [44], four machine-learning algorithms—logistic regression (LR), random forest
(RF), naïve Bayes (NB), and k-nearest neighbors (KNNs)—were compared to predict employee
attrition. The study used the IBM HR Analytics Employee Attrition dataset for model training. To

30
address class imbalance, the synthetic minority oversampling technique (SMOTE) was applied.
The LR algorithm performed the best with SMOTE and weighted class techniques, achieving an
accuracy of 90% [44].
In another study [21], researchers used the IBM HR Analytics Employee Attrition dataset to assess
the effectiveness of various machine-learning methods in predicting staff attrition. The algorithms
were selected following a comprehensive evaluation of all machine-learning algorithms offered by
MATLAB. The study aimed to compare the performance of five selected algorithms—LR, linear
support vector machines (LSVMs), quadratic support vector machines, support vector machines
kernel, and boosted trees—using default MATLAB parameters. The findings revealed that LR
achieved the highest accuracy of 87.78%, making it the best algorithm for forecasting this issue
[21].
In [1], an initial exploratory investigation was proposed to use machine-learning techniques to
predict employee attrition. Five machine-learning models—LR, NB classifier, RF, classification
trees, and a simple neural network (NN) architecture—were used. LR emerged as the most effective
technique, achieving an accuracy of 88% [1].
In [45], the IBM HR Analytics Employee Attrition dataset was used to analyze how objective
factors influence employee attrition and predict employee departure from the company. Various
machine-learning algorithms, including decision tree (DT), Gaussian naïve Bayes (GNB), KNN,
support vector machines (SVMs), RF, LR, and LSVMs, were applied. The GNB algorithm
performed the best, achieving the highest recall rate of 0.54. Additionally, it was noted that the
“Research and Development” department had the highest percentage of resigned workers (133 out
of 237 employees), with an attrition rate of 13.8%, while the “Sales” and “Human Resources”
departments had attrition rates of 20.6% and 19%, respectively [45].
In [20], the objective was to develop a model using supervised classification and machine-learning
techniques to predict voluntary job turnover. The study described various professional and personal
variables that could influence employee attrition rates. The correlation matrix analysis revealed a
positive association between years at the company, years with the current manager, total working
years, and monthly income, while job level, tenure in the current role, total years worked, and
monthly salary were negatively correlated with attrition. Real data from IBM was used to
implement Multinomial NB, GNB, Bernoulli NB, DTs, RFs, and LR algorithms for prediction. The
GNB algorithm, known for its ability to minimize false negatives, achieved the highest recall rate
of 70.76% [20].
In [10], the authors proposed predicting employee attrition rates and evaluating employee emotions
within the company. They included employee attrition-related questions and the Goldberg
Depression Questionnaire in their survey. Various graphs were used to establish connections
between different variables, and machine-learning algorithms such as KNN, DT, SVM, LR, RF,
and GNB were employed. The RF model emerged as the top performer, achieving an accuracy of
86% [10].
Similarly, in a study [40] that utilized the IBM HR Analytics Employee Attrition dataset,
researchers compared the performance of various classification models, such as DT, RF, XGBoost
(XGB), LR, and SVM, in predicting employee attrition. The RF classifier-based model showed
superior predictive capability, achieving an accuracy score of 97%. Furthermore, the Chi-square
test identified several factors that significantly impact an organization’s profit levels, including
overtime, marital status, monthly income, job role, age, total working years, distance from home,
stock options level, job satisfaction, and years at the company [40].

31
In [43], researchers used a real HR dataset from Kazakhstani higher education institutions. The
dataset included demographic, job characteristic, gender, and other relevant data from 2023. It
consisted of approximately 1500 observations, with 39% of them representing cases of employee
attrition. Machine-learning models, including LR, KNN, Multinomial NB, GNB, SVC, and DT,
were applied. However, the initial model predictions were considered insufficiently accurate,
leading to the use of optimization techniques such as Lasso regression. After optimization, DT
emerged as the most accurate model, achieving a score of 95% [43].
In another study [46], the authors focused on systematically predicting attrition using Machine
Learning and Data Analysis methods. They utilized the IBM HR Employee Analytics Attrition and
Performance dataset. To address the imbalance in the dataset, SMOTE was employed to generate
entries for the minority class. Various machine-learning techniques, including DTs, SVMs, KNN,
RF, and NB, were applied. The RF classifier, with feature reduction, achieved an accuracy score
of 83.3%, while the DT classifier reached 81%. This indicates the superiority of these classifiers
over the others [46].
Additionally, a framework for predicting employee attrition using an LR model was presented in
[2]. By utilizing the IBM HR Analytics Employee Attrition dataset, the model’s accuracy improved
from 78% to 81% with feature selection, which highlights its effectiveness [2]. Table 1 shows a
summary of previous studies that applied the machine learning approach, highlighting the results,
strengths, and limitations of each study.
Table 1. Summary of previous studies that applied the machine learning approach.
Ref Date Dataset Used Models Result Strengths Limitations
[44] 2023 IBM HR
Analytics
Employee
Attrition
dataset
LR, RF, NB,
and KNNs
LR performed
the best,
accuracy =
90%
Results were
improved by
addressing the
class
imbalance
(weighted
classes and
SMOTE)
The dataset
used is limited
and does not
account for
every
psychological
and subjective
factor.
[21] 2022 LR, LSVMs,
SVM,
boosted tree
LR achieved
the highest
accuracy
=87.78%.
MATLAB
Classification
Learner was
applied to test
19 ML
algorithms in
6 iterations.
Limited dataset.
All factors were
Taken into
consideration
without filtering
them based on
the most
important.
[1] 2022 LR, NB, RF,
classification
trees, and NN
LR performed
the best
achieving an
accuracy=88%
The main
employee
attrition
reasons were
analyzed.
Limited dataset.
And the data
imbalance was
not handled.
[20] 2021 NB, GNB,
Bernoulli
NB, DTs,
RFs, and LR
The GNB
algorithm
achieved the
highest recall
rate of 70.76%
The
correlation
between
features was
Analyzed.

32
[45] 2020 DT, GNB,
KNN, SVMs,
RF, LR,
LSVMs,
GNB
performed the
best,
achieving the
highest recall
= 0.54
Several basic
indicators of
employee
attrition were
Identified.
Limited dataset.
Feature
engineering was
not applied.
[10] 2021 The dataset
collected by
a
questionnaire
KNN, DT,
SVM, LR,
RF, and GNB
RF emerged
as the top,
achieving an
accuracy=86%
Real dataset,
including the
Goldberg
Depression
Questionnaire.
Improved
methods can be
applied to make
models more
accurate.
[43] 2023 real HR
dataset from
Kazakhstani
higher
education
institutions
LR, KNN,
Multinomial
NB, GNB,
SVC, and DT
DT emerged
as the most
accurate
model,
achieving a
score of 95%.
A real HR
dataset.
Various
optimization
techniques
were
employed to
enhance the
performance
of the model
like Lasso
regression
.
The features of
employee
attrition were
not analyzed.
[46] 2019 IBM HR
Analytics
Employee
Attrition
dataset
DTs, SVMs,
KNN, RF,
and NB
RF, with
feature
reduction,
achieved an
accuracy =
83.3%, while
the DT
reached 81%.
This indicates
the superiority
of these
classifiers
over the others
SMOTE was
applied to
handle data
imbalance.
Feature
Selection
techniques
were
performed to
improve
prediction
performance.
The dataset
used is limited
and does not
account for
every
psychological
and subjective
factor.
[2] 2021 LR the model’s
accuracy
improved
from 78% to
81% with
feature
selection
The max-out
feature
selection
technique was
applied.
Analysis of
the most
important
features of
employee
attrition.
[40] 2022 DT, RF,
XGBoost
(XGB), LR,
and SVM
RF showed
superior
predictive
capability,
Variables that
had a
significant
effect on
Limited dataset.
Improved
methods can be
applied to make

33
achieving an
accuracy=97%
capture were
recorded after
the Chi-square
test.
models more
accurate.
4.2.2. Deep learning
The dataset used for analysis in [30] was the IBM HR Analytics Employee Attrition dataset.
Machine-learning techniques and NNs were employed to analyze the data. Correlation analysis
was used to identify the features that affect employee attrition. Different machine-learning models
such as LR, RF, DT, NB, SVM, multilayer perceptron (MLP), and DNN were compared. RF
performed the best with an accuracy of 97%. Although RFs proved to be the most effective, the
article also demonstrated the promising accuracy of NNs, with an accuracy rate close to 95%.
Initially, the NN model provided 90% accuracy, which improved to approximately 95% [30].
Factor analysis revealed that employees who lived far from their workplace had a higher attrition
rate. In addition, lower-paid workers exhibited higher employee attrition compared to their higher-
paid counterparts. Employee turnover was higher for those involved in more projects compared to
those involved in fewer projects. Workers with more than five years of experience typically stayed
in their jobs longer, and higher-level personnel had lower attrition rates. Workers with higher
education tended to leave the organization more quickly. However, the natural environment was
not found to be a significant factor in attrition analysis [30].
The researchers used the IBM HR Analytics Employee Attrition dataset in [22] to identify
individual traits and structural factors that contribute to employee attrition. They compared various
categorization models, including NB, LR, KNNs, and NNs (MLP). Despite the limited available
data, the NN produced satisfactory results. However, it was noted that the NN model might be
overfitted, which could affect generalization. They explored several hyperparameter
configurations, with the best accuracy reaching 88.43% [22].
In another study [6], because the IBM HR Analytics Employee Attrition dataset was imbalanced
and biased towards employed individuals, the authors used a balanced version obtained through
the adaptive synthetic sampling method. They used a correlation matrix to evaluate the
relationships between the features of the dataset. Deep neural networks (DNNs) were used as the
model and achieved an accuracy of 94% [6].
In another study [8], researchers presented a model for predicting employee attrition using various
predictive analytical techniques, including the artificial neural network (ANN) algorithm, gradient
boosting, RF, and ensemble models. The dataset was sourced from the SAS library (www.sas.com).
Key evaluation metrics, such as average squared error, misclassification rate, and the Gini
coefficient, were used to select the best model, which was determined to be the ANN algorithm
[8].
In [14], an approach using people-analytics to predict employee attrition was introduced, giving
priority to data quality rather than quantity. The study employed three datasets: the IBM HR
Analytics Employee Attrition dataset, a simulated HR dataset from Kaggle, and data gathered
through a questionnaire on the causes of employee attrition. Deep learning algorithms, such as
DNNs, long short-term memory networks, and convolutional neural networks, were utilized,
alongside various machine-learning algorithms and ensemble methods. The study’s findings
indicated that the voting classifier (VC), an ensemble method, outperformed other models across
all datasets [14]. Table 2 shows a summary of previous studies that applied the deep learning
approach, highlighting the results, strengths, and limitations of each study.

34
Table 2. Summary of previous studies that applied the deep learning approach.
[30] 2023 IBM HR
Analytics
Employee
Attrition
dataset
LR, RF, DT,
NB, SVM,
MLP, and
DNN
RF
performed
the best with
an accuracy =
97%
Analysis of
common causes
of employee
turnover rates.
The dataset
used in these
works is
limited. Data
augmentation
techniques
are not
applied.
feature
selection is
not used to
achieve more
accurate
results.
[22] 2022 NB, LR,
KNNs, and
NNs MLP.
The best
accuracy
reached
88.43% for
NN model
Highlighting the
main employee
attrition causes.
[6] 2021 Deep neural
networks
(DNNs)
achieved an
accuracy of
94%
ADASYN is
used to
transform the
original dataset
into a synthetic
one.
Analysis of
employee
attrition features.
[8] 2022 from the SAS
library
ANN,
gradient
boosting, RF,
and ensemble
models,
LSVMs,
The best
model was
ANN.
A real dataset
from SAS was
used.
Employee
attrition
factors were
not analyzed.
feature
selection is
not used to
achieve more
accurate
results.
[14] 2021 Three
datasets: an
IBM dataset,
a dataset
from Kaggle,
and a dataset
gathered by a
questionnaire.
DNNs,
LSTM, and
CNN,
alongside
various ML
algorithms
and ensemble
methods.
The voting
classifier
(VC), an
ensemble
method,
outperformed
other models
across all
datasets
Feature selection
methods were
applied.
Three datasets
were used, and
the prediction
models'
performance in
various
scenarios was
evaluated.
Ignoring the
class-
unbalanced
issue.

35
4.3. Ensemble Methods to Enhance Prediction
The process of systematically creating and merging multiple models, such as classifiers or experts,
to address specific computational intelligence issues is known as ensemble learning. The main
objectives of ensemble learning are to improve a model’s performance (classification, prediction,
function approximation, etc.) or reduce the risk of randomly selecting a poor model [38]. Ensemble
techniques strive to build a set of diverse learning algorithms and combine them to achieve better
prediction performance compared to a single learning algorithm [47].
There are various types of ensemble models depending on how they were trained. Bagging involves
training samples in parallel, using a bootstrap sampling approach with replacement, and then
aggregating the results using a voting method. Boosting, similar to bagging, uses sampling with
replacement, but instead of training in parallel, it learns sequentially. It assigns low weights to
accurate predictions and high weights to incorrect ones, thereby giving more attention to correcting
inaccurate predictions. Lastly, stacking involves generating additional predictions by using the
outputs of each model as input variables. A meta-learner retrains on the data trained by individual
algorithms to make predictions, giving stacking a dual-training characteristic [15].
Most previous research papers compare the accuracy of different models (such as DTs, RFs, SVMs,
etc.) to determine which machine-learning model is best at handling staff attrition. However, a
single model may not achieve the highest level of accuracy due to its development in a specific
setting, which limits its generalizability [15]. Therefore, researchers are interested in ensemble
models that combine multiple models to enhance performance [47]. Many efforts have been made
to improve pattern recognition and prediction on complex and atypical data using machine learning,
which has led some employee attrition prediction studies to focus on applying ensemble models to
improve prediction accuracy. This subsection reviews these studies.
In the study [15], researchers developed a predictive model based on 30 characteristics from the
IBM HR Analytics Employee Attrition dataset that are known to influence employee attrition. They
created eight predictive models, including RF, XGB, SVM, LR, ANN mode, and three different
types of ensemble models that integrate individual models. The goal of the ensemble models was
to combine multiple models into one to outperform any single model alone. By leveraging multiple
models, the ensemble approach aimed to improve forecast accuracy and reduce the overall error
rate. Specifically, the ensemble models included ESM1, which used LR as a meta-learner and RF
and ANN as base algorithms; ESM2, which incorporated LR, ANN, XGB, and SVM as base
algorithms; and ESM3, which combined LR, ANN, XGB, and SVM. Evaluation of the models’
performance in predicting employee attrition revealed that ESM1 achieved the highest accuracy
score of 97% [15].
In another study [23], HR analytics data from the Kaggle website was used to develop a model for
predicting employee turnover rates. Predictive modeling was conducted using the AdaBoost
classifier and the RF classifier. The findings showed that the RF classifier achieved an accuracy of
100%. This achievement can be attributed to the bagging strategy applied by the RF classifier,
which involves dividing the dataset into separate random subsets and randomly selecting features
during each stage of tree splitting. This approach effectively addresses the problem of overfitting
[23].
In the study [42], the researchers investigated the correlation between employee performance and
organizational growth. They collected real-time employee data from a reputable multinational
corporation in Chennai, which consisted of around 5000 employees from different departments,
age groups, and locations. The researchers used the extreme gradient boosting ensemble classifier,
a boosting algorithm, to predict employee performance and then compared it with the gradient

36
boosting ensemble classifier. The results indicated that both algorithms achieved an accuracy rate
of over 90%, with the extreme gradient boosting method outperforming the traditional gradient
boosting ensemble learning approach [42].
To accurately identify employee attrition using machine learning, the researchers utilized the IBM
HR Analytics Employee Attrition dataset in their study [47]. They created six different training
sets, each with a different number of features based on the associations between variables. They
applied several machine-learning algorithms, such as DTs, AdaBoost, LR, RF, gradient boosting,
and ensemble methods. The best-performing ensemble consisted of DT and LR, achieving an
accuracy rate of 86% [47].
In [48], machine-learning algorithms were used to analyze the factors that contribute to employee
attrition and predict attrition itself. The IBM HR Analytics Employee Attrition dataset was used in
this study. An optimized approach, using the extra trees classifier (ETC) algorithm, was applied.
Employee exploratory data analysis was conducted, and in order to balance the dataset, SMOTE
dataset resampling was used. The ETC algorithm achieved the highest accuracy score of 93%,
surpassing SVM, LR, and DT in this experiment [48].
The researchers in [49] used the IBM HR Analytics Employee Attrition dataset to predict the
employee attrition rate using machine-learning algorithms. The study’s findings showed that XGB
performed the best, achieving an accuracy of 88%, when compared to the other algorithms trained
in this study: DT, RF, KNN, NNs, and AdaBoosting [49].
In the study [50], a tree-based ensemble machine-learning model is used with the IBM HR
Analytics Employee Attrition Performance dataset to conduct a comprehensive analysis of
employee attrition. The study aims to optimize the effectiveness of the currently available tree
approaches by evaluating the tree-based ensemble. The machine-learning models employed are RF
and gradient boosting. The kernel density estimate plot is utilized to illustrate the distribution of
data across the features of the dataset and to determine the density of data across different features.
The gradient boost method, an ensemble method, outperforms the RF algorithm as it achieves a
balance between accuracy and recall scores while also improving accuracy. This ensemble
methodology achieves an accuracy of approximately 95%, which is significantly higher than the
RF approach [50].
The model for predicting employee attrition was developed, and the study by [4] identified the
most significant factors that contribute to employee attrition. The dataset used included data from
the human resources department of an Iranian pharmaceutical company. A gradient boosting
algorithm was applied, resulting in an accuracy rate of 89% [4].
In [51], the authors developed a method to assess the importance of each factor that impacts an
employee’s likelihood of leaving the organization, using the IBM HR Analytics Employee Attrition
dataset. The following machine-learning algorithms were employed: AdaBoost, XGB, gradient
boosting, KNN, DT, RF, and LR. After testing, evaluation, and validation, it was determined that
the XGB classifier yielded the most favorable outcomes, achieving an accuracy of 87% [51].
A people-analytics approach for predicting employee attrition was proposed in [14]. The study
focused on data quality rather than quantity. Three datasets were used: the IBM HR Analytics
Employee Attrition dataset, a simulated HR dataset supported by Kaggle data, and a dataset
collected through a questionnaire about the causes of employee attrition. Machine-learning
algorithms like DT, SVM, LR, and RF, along with ensemble methods like XGB, VC, and stacked
ANN-based, were applied, as well as a set of deep learning algorithms. The study’s results reveal
that VC outperforms the other models on all datasets [14].

37
The ensemble model showed superiority over the rest of the models, in both studies [52] and [53].
The attrition rate of employees is discussed by the authors in [52]. To benchmark employee attrition
accuracy, three distinct supervised learning algorithms AdaBoost, SVM, and Random Forest are
used. The models were trained using the IBM dataset. Most of the time, AdaBoost has
outperformed the other classifiers in terms of accuracy. The accuracy gained from several
algorithm executions on the random splitting of training and test data was in the range of 0.84 to
0.88 when there were 1000 estimators and a learning rate of 0.1 [52].
The authors in [53] suggest a technique for creating a turnover intention prediction model based on
machine learning. An analysis was done in this study to confirm the influence of predictors using
data from the Job Movement Path Survey for college graduates conducted by the Korea
Employment Information Service. Extreme gradient boosting (XGB), logistic regression (LR), and
k-nearest neighbor (KNN) classifiers were used for model learning and classification. Looking at
the average values of the four sets of cross-validation, the turnover intention prediction model
analysis revealed that XGB scored the highest accuracy (78.5%), followed by LR (78.3) and KNN
(76%). One of the model benefits in this study is calculating the significance of independent
variable input to the prediction model. The significance ranking of the independent variables was
examined in this investigation. The research revealed that contentment with work security was the
most significant predictor in predicting the desire to leave one's employment. Following that,
contentment with the organization and the work's alignment with the employee study major
emerged sequentially [53].
The authors in [41] describe and analyze a special study on employee turnover intention conducted
across Europe. The study is based on the GEEI survey carried out by Effectory in 2018. This labor
market survey included a sample of 18,322 employees from 56 European countries. The survey
consisted of 123 questions covering contextual information such as socio-demographic, labor, and
industry aspects, as well as an objective question. A subset of the raw data from the GEEI survey,
containing 9296 rows, was selected and limited to respondents from 30 European countries. The
predictive abilities of various machine-learning classifiers, including KNN, LR, DT, RFs,
LightGBM (LGBM), XGB, and TabNet, were compared. The top-performing models were found
to be LGBM and LR. To determine which features were responsible for the performance of both
LR and LGBM, the authors conducted an analysis of the relevance of each feature by performing
100 separate trials and compiling importance ratings from all trials. The absolute value of the
coefficient for a feature in the LR model was used as a measure of its significance, while the
relevance of a feature in the LGBM model was determined by how frequently it was used in the
model tree split [41]. Table 3 shows a summary of previous studies that applied ensemble methods,
highlighting the results, strengths, and limitations of each study.

38
Table 3. Summary of previous studies that applied ensemble methods.
[15] 2022 IBM HR
Analytics
Employee
Attrition dataset
RF, XGB,
SVM, LR,
ANN mode,
and three
different
types of
ensemble
models that
integrate
individual
models
ESM1
achieved the
highest
accuracy
score of 97%
Multiple
approaches were
applied,
including ML,
deep learning,
and a
combination of
ensemble
models.
SMOTE
technique was
performed.
The dataset
used is limited
and does not
account for
every
psychological
and subjective
factor.
[49] 2022 XGB and
DT, RF,
KNN, NNs,
and
AdaBoosting
XGB
performed the
best,
achieving an
accuracy of
88%
Several feature
selection
methods were
used.
[48] 2022 extra trees
classifier
(ETC)
algorithm
, SVM, LR,
and DT
The ETC
algorithm
achieved the
highest
accuracy
score of 93%
The dataset of
employee
attrition was
analyzed by the
EEDA.
SMOTE
technique was
applied.
[47] 2021 DTs,
AdaBoost,
LR, RF,
gradient
boosting, and
ensemble
methods
The best-
performing
ensemble
consisted of
DT and LR,
achieving an
accuracy rate
of 86%.
Multiple models
in several
scenarios were
applied to
determine the
best model.
Limited
dataset.
Improved
methods can
be applied to
increase the
accuracy rate
and reduce
generalization
errors.
[50] 2021 IBM HR
Analytics
Employee
Attrition dataset
RF and
gradient
boosting
gradient
boosting
achieves an
accuracy of
approximately
95%, which is
significantly
higher than
the RF
approach.
Exploratory Data
Analysis was
Applied to
Feature
Engineering.
The dataset
used is limited
and does not
account for
every
psychological
and subjective
factor.

39
[51] 2022 AdaBoost,
XGB,
gradient
boosting,
KNN, DT,
RF, and LR
XGB yielded
the most
favorable
outcomes,
achieving an
accuracy of
87%
Analysis and
recommendations
about employee
attrition were
provide.
[52] 2024 AdaBoost,
SVM, and
Random
Forest
AdaBoost has
outperformed
the others in
terms of
accuracy.
Accuracy was
tested in three
different random
iterations.
Analysis of
features
correlation was
applied.
[23] 2022 HR analytics
data from the
Kaggle.
AdaBoost
classifier and
the RF
classifier.
RF classifier
achieved an
accuracy of
100%.
Feature
engineering was
applied to select
the relevant data.
Ignore the
dataset
imbalance
problem.
[42] 2022 They collected
real-time
employee data
from a reputable
multinational
corporation in
Chennai.
extreme
gradient
boosting
ensemble
classifier, a
boosting
algorithm.
both
algorithms
achieved an
accuracy rate
of over 90%,
with the XGB
method
outperforming
the traditional
gradient
boosting
ensemble.
The importance
of features was
analyzed, and
then the most
prominent
features were
selected using
chi-square and
mutual
information.
It is possible to
compare the
performance
of the model
with other
ensemble
learning
algorithms.
[4] 2021 data from the
human
resources
department of
an Iranian
pharmaceutical
company
A gradient
boosting
algorithm
A gradient
boosting
algorithm was
applied,
resulting in an
accuracy rate
of 89%
Analysis to
identify
employee
attrition factors
was performed.
The valuation
of experiments
is limited to
the
pharmaceutical
sector.
[53] 2024 Dataset from the
Job Movement
Path Survey for
college
graduates
conducted by
the Korea
XGB, LR,
and KNN
XGB had the
highest
accuracy
(78.5%),
The reasons for
choosing a job
were examined,
and
recommendations
to organizations
for managing
new talent were
provided.
The limited
application of
classification
techniques,
and the data
imbalance in
turnover
intention.

40
[41] 2022 The study is
based on the
GEEI survey
carried out by
Effectory in
2018
KNN, LR,
DT, RFs,
LGBM,
XGB, and
TabNet,
The top-
performing
models were
found to be
LGBM and
LR
The study
covered a dataset
covering 30
European
countries.
The study
combined
analytical
methodologies
from the
perspective of
predictive
analysis of
human resources
with a causal
approach.
The limitation
is the
weighing of
datasets to
combat
selection bias.
which is
restricted to
the workforce
of a single
nation.
4.4. Feature Selection Techniques
Feature selection techniques in machine learning aim to find the optimal set of features for
constructing optimized prediction models [3]. These techniques eliminate unnecessary and
redundant attributes and select relevant features to improve accuracy [14].
Several feature selection techniques have been developed for this purpose, including filter,
wrapper, embedded, and hybrid approaches. Filtering techniques select relevant features based on
the correlation between feature values. Wrapper approaches choose features based on their
performance with a selected subset of features. Embedded approaches incorporate feature selection
directly into the learning process. Hybrid methods combine elements from these approaches to
achieve optimal feature selection [2].
The use of feature selection algorithms in datasets is crucial for improving accuracy. Recent
research has extensively investigated the selection of significant features from data to enhance the
speed and accuracy of models [2].
The feature selection process is important because it improves the accuracy of classification The
feature selection process is significant as it improves the accuracy of classification results for the
underlying problem. Real-world datasets typically contain a large number of features, many of
which may not be relevant to classification or prediction models. This abundance of unnecessary
and redundant features can result in inconsistent and unreliable results when all features are used
in classification or prediction tasks [37]. Therefore, feature selection plays a critical role in
identifying attributes that genuinely contribute to classification accuracy, especially those closely
related to the class variables [38].
In this subsection, employee attrition prediction studies are reviewed regarding their reliance on
feature selection algorithms.
Since employee datasets contain numerous features, some of which may be irrelevant, researchers
often choose to select features that are relevant to the issue of employee attrition. For example, in
studies like [47] and [50], principal component analysis (PCA) was used to reduce the
dimensionality of the feature space. However, PCA may struggle to handle binary variables that
are common in attrition databases. Additionally, PCA relies on linear relationships between

41
features for dimension reduction and may overlook potential connections between the output and
filter features [2].
Researchers aim to identify the features that have the greatest impact on employee attrition by
employing feature selection techniques that reduce or eliminate unnecessary features. In many
experiments, researchers applied feature selection techniques and compared the results. Notably,
the RF classifier technique for feature selection has proven effective in improving RF metrics. For
example, the accuracy of the RF classifier increased from 85.22% to 86.10% [23].
In [3], the IBM HR Analytics Employee Attrition dataset was used to conduct RF classification
modeling to predict employee attrition. The study focuses on comparing feature selection
techniques to eliminate or reduce redundant characteristics, to identify features that have the
greatest impact on employee turnover. Three feature selection algorithms were used: SelectKBest,
information gain, and recursive feature elimination. According to the test scenario in this study,
information gain achieved the highest accuracy value of 89.2% when utilizing 25 features [3].
In [38], a model was presented for predicting employee attrition using an enhanced weight-based
forest optimization algorithm. Among all the sets of features, this algorithm chooses the best one.
A new classifier, called improved random forest (IRF), was proposed, which selects n samples
from a total of k samples to build a random number of DTs. The final prediction is assessed based
on the performance of the classifier rather than a majority vote. The analysis was performed using
the IBM HR Analytics Employee Attrition dataset. The study’s findings show a model accuracy of
91% [38].
In [14], the feature selection techniques RFE and SelectKBest were used. Features selected by
SelectKBest are kept even if they are eliminated by RFE. Similarly, features not eliminated by RFE
but selected by SelectKBest are also kept. On the other hand, features rejected by RFE and not
selected by SelectKBest are discarded. As a result, age, marital status, rewards, job involvement,
training, business travel, tenure, grade, job performance, job satisfaction, and environment
satisfaction are identified as the 11 main features related to employee attrition prediction, based on
the combination of the RFE and SelectKBest feature selection techniques [14].
In [49], after applying the chi-square, recursive method, mutual information method, and tree
selector methods to each attrition-related feature, only the top 15 elements from each approach
were combined into a set, resulting in a consideration of 23 features. Additionally, eight PCA
characteristics were included, resulting in a total of 31 features being considered [49].
In another experiment [42], mutual information and chi-square were two alternative methods used
for feature selection. Features showing the highest mutual information and chi-square values with
the class label were selected [42].
In several experiments, various techniques for selecting features were implemented, contributing
to the improvement of prediction results. The Boruta method was used to identify features that
influence employee attrition and rank them based on their average importance [4]. Additionally,
the “max-out” feature method was applied to minimize the difference between features [2].
Exploratory data analysis was conducted in [51] to gain a better understanding of the dataset, using
a variety of graphs to visualize employee attrition more comprehensively. Furthermore, the Shapley
additive explanations index was used to determine the most significant factors contributing to the
attrition [51].

42
5. DISCUSSION
General discussion: The study included a literature survey of previous studies that presented an
attempt to predict employee attrition from 2019 to February 2024. Most previous studies took a
machine learning approach to predict employee attrition, although a group of them relied on the
machine learning approach. Most studies agree to conduct comparative experiments for several
predictive models and then determine the model that performs best in predicting employee attrition.
Discussion related to algorithms: The most widely used machine learning algorithm in employee
attrition prediction is linear regression. In most cases, the linear regression algorithm is used to
compare with the study's suggested algorithms and verify the algorithm with the best performance,
as in studies [10], [20], and others. However, its appearance in many studies does not mean that it
is the best-performing model for predicting employee attrition. Through this literature survey, the
Random Forest algorithm can be considered the best-performing algorithm, and its superiority over
many algorithms has been repeated, as in [10], [40], and others. The model (RF) considers multiple
factors and has an accuracy of 86.0% in predicting employee attrition, which is greater than the
performance reported in [10]. According to the chosen performance metrics (Accuracy, Precision,
Recall, and F1-score), the RF classifier-based model is considered to be the best. The model
achieved an accuracy of 97.8% [40]. The RF model is the most promising in terms of performance
metrics considered, compared to other models in the researchers' experiment [1], including the
Neural Network model. The RF model achieved an accuracy of 87.96%, while the NN model
achieved an accuracy of 84.76%. As a result, applying RF not only enhances data-driven decision-
making but also yields plausible justifications for choices made. The last point is particularly
significant in real-world circumstances where the outcomes' justice and dependability are
paramount [1]. In addition, the turnover intention prediction analyses revealed that the XGB
ensemble model outperforms many other models.
A recent use of deep learning (DL), a subfield of machine learning, for employee attrition
prediction issues shows great promise. We also found some deep learning-based studies in this
investigation. DL approaches have various potential qualities, like better performance and
automated feature extraction. We anticipate that further studies on the application of DL techniques
in employee attrition prediction will be carried out in the future.
Discussion related to datasets: A variety of datasets were utilized to study the issue of employee
attrition and develop predictive models using them. Some studies relied on ready-made datasets
available or archival employee datasets for specific institutions, and some researchers collected
data using questionnaires that included questions about factors related to employee attrition.
However, we can say that the available dataset (IBM HR Analytics Employee Attrition dataset)
produced by IBM, which includes 1470 observations, is the most widely used in previous studies,
as is the case in the studies of [15], [22], [23], [40], and many others.
This dataset is highly unbalanced, with 237 positive samples (from former employees) and 1,233
negative samples (from current employees), which leads to a significant data bias [6]. So we can
conclude that without addressing the class imbalance in the data, most models will not perform
well [44]. Addressing dataset imbalance using the Synthetic Minority Oversampling (SMOTE)
technique and weighted classes has improved prediction results as in [44]. When paired with
weighted class and SMOTE approaches, Logistic Regression yields the maximum recall score of
0.9014 and can accurately predict 64 out of 71 minority class situations. As a result, this study uses
SMOTE and category weights to get the best results from the logistic regression model. Also,
SMOTE is a strategy for oversampling the minority class; it was chosen over undersampling
because the latter could lead to the omission of important data [46]. Also, this dataset is limited.
Therefore, it is recommended to apply data augmentation techniques in the future.

43
On the other hand, we can conclude that the reliance of many research studies on the same dataset
is an indication of the lack of ready and available datasets that include many advantages that
contribute to predicting employee attrition with greater accuracy. It is recommended to create real
datasets in the future, which in turn will help organizations examine the factors that lead to
employee attrition, especially if they focus on a specific field or sector.
Discussion related to features: Some studies have taken all features of the database into account
without analyzing and filtering them based on their importance. While other studies provided
analyses to determine the most important factors that cause employee attrition.
To name a few, the outcome obtained in [45] shows that the main factors of attrition are monthly
income, age, overtime, and distance from home. The results of the analysis in [20] revealed that
there is a positive relationship between employee attrition and the following characteristics: years
working in the company, years with the current manager, total years of work, and monthly income.
There is also a negative relationship between attrition and job level, years in a current role, total
years of work, and monthly income [20]. In [41] can be noted that time spent with the company,
satisfactory development, and willingness to recommend the employer to friends and family are
among the top five traits in the study. It's interesting to note that in the two data sets included in
the study, gender—a widely acknowledged driver of turnover intention—is not ranked among the
most significant characteristics. It also does not show any country-specific effect [41].
Therefore, the most prominent features whose importance in prediction was praised by previous
studies were summarized.
The most important features affecting employee attrition that were extracted are shown in Table 4.
The number of times these features are mentioned is displayed, along with references. As shown
in Table 4, the most influential features are monthly income, age, years of experience, overtime,
job role, environmental satisfaction, and job satisfaction. Depression level, bonuses, training,
business travel, hourly wage, gender, department, last contract length, childcare allowance, and
working on multiple projects were among the features that were not significant in most studies.
Table 4. Summary of the features that studies have proven to affect employee attrition.
Factors Times References
Monthly Income 11 [1], [4], [6], [10], [15], [23], [30], [40], [45], [48],[51].
Age 10 [1], [4], [14], [15], [23], [38], [40], [45], [48],[51].
Years of Experience 8 [2], [4], [23], [30], [38], [40], [45], [48].
Overtime 7 [2], [6], [15], [38], [40], [45], [51].
Job Role 6 [2], [4], [6], [14], [40], [51].
Environment Satisfaction 6 [1], [14], [15], [38], [45], [51].
Job Satisfaction 6 [10], [14], [23], [38], [40], [51].
Job Involvement 5 [1], [2], [14], [30], [38].
Distance from Home 5 [1], [30], [40], [45], [51].
Marital Status 4 [14], [23], [38], [40].
Number companies worked 4 [2], [15], [23], [51].
Work-Life Balance 3 [1], [23], [38].
Stock Option Level 3 [40], [45], [51].
Years with current manager 3 [2], [23], [38].
Years in the company 3 [1], [40], [45].
Job Performance 2 [14], [15].
Relationship Satisfaction 2 [15], [51].

44
Direct manager 2 [4], [48].
Grade 2 [4], [14].
Years after the last promotion 1 [2].
Depression Level 1 [10].
Rewards 1 [14].
Training 1 [14].
Business Travel 1 [14].
Hourly rate 1 [48].
Gender 1 [15].
Department 1 [4].
Last Contract Duration 1 [4].
Allowance for Childcare 1 [4].
Work on multiple projects 1 [30].
As for the practical implications based on the results of this study, human resources management
can use machine learning models to forecast and analyze employee attrition behavior and, in turn,
develop plans to increase job satisfaction. Additionally, managers can lessen the causes of attrition
behaviors within their company and offer the best ways to eliminate them.
The survey found several significant reasons why workers quit their jobs. It is possible to design
effective policies and initiatives that will support employee retention by taking these aspects into
account. The characteristics that have a major impact on employee attrition are monthly salary,
age, years of experience, extra time, job role, environmental satisfaction, and job satisfaction, as
has been repeatedly stated in prior research. Therefore, it can be suggested to include the following
policies in the strategic plans:
- Competitive base pay.
- Reduction in the "number of projects/tasks assigned" to workers who are working
beyond their capacity.
- Regular promotions, rewards, and incentives
- Motivate staff members and encourage them to work only during business hours.
- Recognizing talents and offering flexibility.
- providing a suitable work environment, providing the required opportunities and
resources.
- Ensuring the satisfaction of employees to continue in the organization and thus
reducing the “Employee Attrition” [16] [38].
6. CONCLUSIONS
This study showed that the selected publications used various machine learning models along with
multiple feature selection methods, depending on data availability. Each study looks at predicting
employee attrition using machine learning, the various features used, and the type and size of the
database. Some studies relied on ready datasets available on the Internet, while in other studies,
researchers collected data for their studies.
Many models have been used in different studies. However, it cannot be considered that any
specific model is the best model ever to predict employee attrition. However, the Random Forest
model has shown superior performance in several studies. The most used models in previous
studies were Linear Regression, Decision Trees, Random Forests, and Support Vector Machines.
Most studies relied on comparing a group of different machine learning models to determine the
best-performing model among them. It has been observed that the ANN and DNN algorithms are

45
the most preferred deep learning algorithms. This article can be considered that pave the way for
further research on developing the problem of crop yield forecasting.
In future work, we aim to build on the results of this study and focus on developing a DL-based
employee attrition prediction model. It is also recommended to focus on improving the
performance of models by implementing modern feature selection methods, which contribute to
identifying the factors that most affect prediction accuracy and thus excluding unnecessary
features.
Numerous research papers' reliance on the same dataset suggests that there aren't enough readily
available datasets with a variety of factors that help forecast employee attrition more accurately. It
is recommended to create real data sets in the future, that are collected based on the analysis of
factors that lead to employee attrition, especially if they focus on a specific area or sector of society.
Since the process of predicting the departure of an employee is a critical decision that requires trust
in the machine learning model, and at the same time, decision-makers in the human resources
department are keen to know the reasons behind the departure of employees, it is recommended to
apply explainable artificial intelligence.
REFERENCES
[1] F. Guerranti and G. M. Dimitri, “A Comparison of Machine Learning Approaches for Predicting
Employee Attrition,”Applied Sciences, vol. 13, no. 1, p. 267, Dec. 2022, doi: 10.3390/app13010267.
[2] S. Najafi-Zangeneh, N. Shams-Gharneh, A. Arjomandi-Nezhad, and S. Hashemkhani Zolfani, “An
Improved Machine Learning-Based Employees Attrition Prediction Framework with Emphasis on
Feature Selection,”Mathematics, vol. 9, no. 11, p. 1226, May 2021, doi: 10.3390/math9111226.
[3] S. F. Sari and K. M. Lhaksmana, “Employee Attrition Prediction Using Feature Selection with
Information Gain and Random Forest Classification,”JoSYC, vol. 3, no. 4, pp. 410–419, Sep. 2022,
doi: 10.47065/josyc.v3i4.2099.
[4] F. Mozaffari, M. Rahimi, H. Yazdani, and B. Sohrabi, “Employee attrition prediction in a
pharmaceutical company using both machine learning approach and qualitative data,”BIJ, vol. 30, no.
10, pp. 4140–4173, Dec. 2023, doi: 10.1108/BIJ-11-2021-0664.
[5] N. Alshanbri, M. Khalfan, M. A. Noor, D. Dutta, K. Zhang, and T. Maqsood, “Employees’ Turnover,
Knowledge Management and Human Recourse Management: A Case of Nitaqat Program,”IJSSH, vol.
5, no. 8, pp. 701–706, 2015, doi: 10.7763/IJSSH.2015.V5.543.
[6] S. Al-Darraji, D. G. Honi, F. Fallucchi, A. I. Abdulsada, R. Giuliano, and H. A. Abdulmalik, “Employee
Attrition Prediction Using Deep Neural Networks,”Computers, vol. 10, no. 11, p. 141, Nov. 2021, doi:
10.3390/computers10110141.
[7] J. Farrugia, “7 statistics on employee turnover every HR manager should be aware of | Workforce.com.”
Accessed: Feb. 19, 2024. [Online]. Available: https://workforce.com/news/7-statistics-on-employee-
turnover-in-2022-every-hr-manager-should-be-aware-of
[8] F. K. Alsheref, I. E. Fattoh, and W. M.Ead, “Automated Prediction of Employee Attrition Using
Ensemble Model Based on Machine Learning Algorithms,”Computational Intelligence and
Neuroscience, vol. 2022, p. 7728668, Jun. 2022, doi: 10.1155/2022/7728668.
[9] Abdelmohsen A. Nassani, Shaykah A. Al-Hassan, and Anhar A. Al-Dakhil, “The Nexus between Work
Engagement, Job Stress, and Employee Turnover Intention in Riyadh Private Sector,” 2021, [Online].
Available: https://doi.org/10.15520/jassh.v7i4.611
[10] R. Joseph, S. Udupa, S. Jangale, K. Kotkar, and P. Pawar, “Employee Attrition Using Machine
Learning And Depression Analysis,” in 2021 5th International Conference on Intelligent Computing
and Control Systems (ICICCS), Madurai, India: IEEE, May 2021, pp. 1000–1005. doi:
10.1109/ICICCS51141.2021.9432259.
[11] P. K. Jain, M. Jain, and R. Pamula, “Explaining and predicting employees’ attrition: a machine learning
approach,”SN Applied Sciences, vol. 2, no. 4, p. 757, Mar. 2020, doi: 10.1007/s42452-020-2519-4.
[12] G. Gabrani and A. Kwatra, “Machine Learning Based Predictive Model for Risk Assessment of
Employee Attrition,” in Computational Science and Its Applications – ICCSA 2018, vol. 10963, O.
Gervasi, B. Murgante, S. Misra, E. Stankova, C. M. Torre, A. M. A. C. Rocha, D. Taniar, B. O.

46
Apduhan, E. Tarantino, and Y. Ryu, Eds., in Lecture Notes in Computer Science, vol. 10963. , Cham:
Springer International Publishing, 2018, pp. 189–201. doi: 10.1007/978-3-319-95171-3_16.
[13] E. Rombaut and M.-A. Guerry, “Predicting voluntary turnover through human resources database
analysis,”MRR, vol. 41, no. 1, pp. 96–112, Jan. 2018, doi: 10.1108/MRR-04-2017-0098.
[14] N. B. Yahia, J. Hlel, and R. Colomo-Palacios, “From Big Data to Deep Data to Support People
Analytics for Employee Attrition Prediction,”IEEE Access, vol. 9, pp. 60447–60458, 2021, doi:
10.1109/ACCESS.2021.3074559.
[15] D. Chung, J. Yun, J. Lee, and Y. Jeon, “Predictive model of employee attrition based on stacking
ensemble learning,”Expert Systems with Applications, vol. 215, p. 119364, Apr. 2023, doi:
10.1016/j.eswa.2022.119364.
[16] S. N. Khera and Divya, “Predictive Modelling of Employee Turnover in Indian IT Industry Using
Machine Learning Techniques,”Vision, vol. 23, no. 1, pp. 12–21, Mar. 2019, doi:
10.1177/0972262918821221.
[17] R. Punnoose and P. Ajit, “Prediction of Employee Turnover in Organizations using Machine Learning
Algorithms,”ijarai, vol. 5, no. 9, 2016, doi: 10.14569/IJARAI.2016.050904.
[18] D. S. Sisodia, S. Vishwakarma, and A. Pujahari, “Evaluation of machine learning models for employee
churn prediction,” in 2017 International Conference on Inventive Computing and Informatics (ICICI),
Coimbatore: IEEE, Nov. 2017, pp. 1016–1020. doi: 10.1109/ICICI.2017.8365293.
[19] S. S. Alduayj and K. Rajpoot, “Predicting Employee Attrition using Machine Learning,” in 2018
International Conference on Innovations in Information Technology (IIT), Nov. 2018, pp. 93–98. doi:
10.1109/INNOVATIONS.2018.8605976.
[20] A. Habous, E. H. Nfaoui, and Y. Oubenaalla, “Predicting Employee Attrition using Supervised
Learning Classification Models,” in 2021 Fifth International Conference On Intelligent Computing in
Data Sciences (ICDS), Oct. 2021, pp. 1–5. doi: 10.1109/ICDS53782.2021.9626761.
[21] N. Khalifa, M. Alnasheet, and H. Kadhem, “Evaluating Machine Learning Algorithms to Detect
Employees’ Attrition,” in 2022 3rd International Conference on Artificial Intelligence, Robotics and
Control (AIRC), Cairo, Egypt: IEEE, May 2022, pp. 93–97. doi: 10.1109/AIRC56195.2022.9836981.
[22] G. Raja Rajeswari., R. Murugesan, R. Aruna., B. Jayakrishnan, and K. Nilavathy., “Predicting
Employee Attrition through Machine Learning,” in 2022 3rd International Conference on Smart
Electronics and Communication (ICOSEC), Trichy, India: IEEE, Oct. 2022, pp. 1370–1379. doi:
10.1109/ICOSEC54921.2022.9952020.
[23] S. Krishna and S. Sidharth, “Analyzing Employee Attrition Using Machine Learning: the New AI
Approach,” in 2022 IEEE 7th International conference for Convergence in Technology (I2CT),
Mumbai, India: IEEE, Apr. 2022, pp. 1–14. doi: 10.1109/I2CT54291.2022.9825342.
[24] F. H. Wardhani and K. M. Lhaksmana, “Predicting Employee Attrition Using Logistic Regression With
Feature Selection,”SinkrOn, vol. 7, no. 4, pp. 2214–2222, Oct. 2022, doi:
10.33395/sinkron.v7i4.11783.
[25] Sami M Abbasi and Kenneth W Hollman, “Turnover: The Real Bottom Line,”Public Personnel
Management, vol. 29, no. 3, pp. 333–342, Sep. 2000, doi: 10.1177/009102600002900303.
[26] A. Mhatre, A. Mahalingam, M. Narayanan, A. Nair, and S. Jaju, “Predicting Employee Attrition along
with Identifying High Risk Employees using Big Data and Machine Learning,” in 2020 2nd
International Conference on Advances in Computing, Communication Control and Networking
(ICACCCN), Greater Noida, India: IEEE, Dec. 2020, pp. 269–276. doi:
10.1109/ICACCCN51052.2020.9362933.
[27] H. Ongori, “A review of the literature on employee turnover,” 2007, [Online]. Available: Available
online http://www.academicjournals.org/ajbm
[28] s. R. Basariya and R. Ahmed, “A STUDY ON ATTRITION – TURNOVER INTENTIONS OF
EMPLOYEES,”International Journal of Civil Engineering and Technology, vol. 10, pp. 1954–19601,
May 2019.
[29] C. Abdullah Al Mamun and Md. Nazmul Hasan, “Factors affecting employee turnover and sound
retention strategies in business organization: a conceptual view,”Problems and Perspectives in
Management, vol. 15, no. 1, pp. 63–71, Mar. 2017, doi: 10.21511/ppm.15(1).2017.06.
[30] S. R. Pulari, A. Punitha, S. Raja Varshni Meenachi, and S. Vasudevan, “A Comparative Study of
Employee Attrition Analysis Using Machine Learning and Deep Learning Techniques,” in Inventive
Communication and Computational Technologies, G. Ranganathan, X. Fernando, and Á. Rocha, Eds.,
Singapore: Springer Nature Singapore, 2023, pp. 1–12.

47
[31] S. Chatterjee, “Understanding the Ergonomics of Attrition,”European Journal of Studies in
Management and Business, vol. 22, pp. 11–18, Sep. 2022, doi: 10.32038/mbrq.2022.22.02.
[32] K. Ashokkumar, S. Jacob, and A. Joseph, “A Study on Attrition Management in Private Sector
Financial Institutions - A Special Reference to Kottayam District in Kerala,”commerce, vol. 8, no. 2,
pp. 35–44, Apr. 2020, doi: 10.34293/commerce.v8i2.2319.
[33] P. Ghosh, R. Satyawadi, J. Prasad Joshi, and Mohd. Shadman, “Who stays with you? Factors predicting
employees’ intention to stay,”Int J of Org Analysis, vol. 21, no. 3, pp. 288–312, Jul. 2013, doi:
10.1108/IJOA-Sep-2011-0511.
[34] M. M. Taye, “Understanding of Machine Learning with Deep Learning: Architectures, Workflow,
Applications and Future Directions,”Computers, vol. 12, no. 5, 2023, doi:
10.3390/computers12050091.
[35] A. Singh, N. Thakur, and A. Sharma, “A review of supervised machine learning algorithms,” in 2016
3rd International Conference on Computing for Sustainable Global Development (INDIACom), Mar.
2016, pp. 1310–1315.
[36] V. V. Saradhi and G. K. Palshikar, “Employee churn prediction,”Expert Systems with Applications,
vol. 38, no. 3, pp. 1999–2006, Mar. 2011, doi: 10.1016/j.eswa.2010.07.134.
[37] U. Moorthy and U. D. Gandhi, “Forest optimization algorithm‐based feature selection using classifier
ensemble,”Computational Intelligence, vol. 36, no. 4, pp. 1445–1462, Nov. 2020, doi:
10.1111/coin.12265.
[38] S. Porkodi, S. Srihari, and N. Vijayakumar, “Talent management by predicting employee attrition using
enhanced weighted forest optimization algorithm with improved random forest classifier,”IJATEE, vol.
9, no. 90, May 2022, doi: 10.19101/IJATEE.2021.875340.
[39] Dr. T. S. Poornappriya and Dr. R. Gopinath, “Employee Attrition In Human Resource Using Machine
Learning Techniques,” 2021.
[40] S. Aggarwal, M. Singh, S. Chauhan, M. Sharma, and D. Jain, “Employee Attrition Prediction Using
Machine Learning Comparative Study,” in Intelligent Manufacturing and Energy Sustainability, vol.
265, A. N. R. Reddy, D. Marla, M. N. Favorskaya, and S. C. Satapathy, Eds., in Smart Innovation,
Systems and Technologies, vol. 265. , Singapore: Springer Singapore, 2022, pp. 453–466. doi:
10.1007/978-981-16-6482-3_45.
[41] M. Lazzari, J. M. Alvarez, and S. Ruggieri, “Predicting and explaining employee turnover
intention,”Int J Data Sci Anal, vol. 14, no. 3, pp. 279–292, Sep. 2022, doi: 10.1007/s41060-022-00329-
w.
[42] P. Sujatha and R. S. Dhivya, “Ensemble Learning Framework to Predict the Employee Performance,”
in 2022 Second International Conference on Power, Control and Computing Technologies (ICPC2T),
Raipur, India: IEEE, Mar. 2022, pp. 1–7. doi: 10.1109/ICPC2T53885.2022.9777078.
[43] B. Meraliyev, A. Karabayeva, T. Altynbekova, and Y. Nematov, “Attrition Rate Measuring In Human
Resource Analytics Using Machine Learning,” in 2023 17th International Conference on Electronics
Computer and Computation (ICECCO), Kaskelen, Kazakhstan: IEEE, Jun. 2023, pp. 1–6. doi:
10.1109/ICECCO58239.2023.10146602.
[44] K. Sheth, J. Patel, and J. Verma, “Machine Learning-Based Investigation of Employee Attrition
Prediction and Analysis,” in Emerging Technology Trends in Electronics, Communication and
Networking, vol. 952, R. Dhavse, V. Kumar, and S. Monteleone, Eds., in Lecture Notes in Electrical
Engineering, vol. 952. , Singapore: Springer Nature Singapore, 2023, pp. 221–238. doi: 10.1007/978-
981-19-6737-5_19.
[45] F. Fallucchi, M. Coladangelo, R. Giuliano, and E. William De Luca, “Predicting Employee Attrition
Using Machine Learning Techniques,”Computers, vol. 9, no. 4, p. 86, Nov. 2020, doi:
10.3390/computers9040086.
[46] N. Bhartiya, S. Jannu, P. Shukla, and R. Chapaneri, “Employee Attrition Prediction Using
Classification Models,” in 2019 IEEE 5th International Conference for Convergence in Technology
(I2CT), Mar. 2019, pp. 1–6. doi: 10.1109/I2CT45611.2019.9033784.
[47] A. Qutub, A. Al-Mehmadi, M. Al-Hssan, R. Aljohani, and H. S. Alghamdi, “Prediction of Employee
Attrition Using Machine Learning and Ensemble Methods,”IJMLC, vol. 11, no. 2, pp. 110–114, Mar.
2021, doi: 10.18178/ijmlc.2021.11.2.1022.
[48] Raza, K. Munir, M. Almutairi, F. Younas, and M. M. S. Fareed, “Predicting Employee Attrition Using
Machine Learning Approaches,”Applied Sciences, vol. 12, no. 13, p. 6424, Jun. 2022, doi:
10.3390/app12136424.

48
[49] L. S. Ganthi, Y. Nallapaneni, D. Perumalsamy, and K. Mahalingam, “Employee Attrition Prediction
Using Machine Learning Algorithms,” in Proceedings of International Conference on Data Science and
Applications, vol. 288, M. Saraswat, S. Roy, C. Chowdhury, and A. H. Gandomi, Eds., in Lecture
Notes in Networks and Systems, vol. 288. , Singapore: Springer Singapore, 2022, pp. 577–596. doi:
10.1007/978-981-16-5120-5_44.
[50] V. Mehta and S. Modi, “Employee Attrition System Using Tree Based Ensemble Method,” in 2021
2nd International Conference on Communication, Computing and Industry 4.0 (C2I4), Dec. 2021, pp.
1–4. doi: 10.1109/C2I454156.2021.9689398.
[51] K. M. Mitravinda and S. Shetty, “Employee Attrition: Prediction, Analysis Of Contributory Factors
And Recommendations For Employee Retention,” in 2022 IEEE International Conference for Women
in Innovation, Technology & Entrepreneurship (ICWITE), Bangalore, India: IEEE, Dec. 2022, pp. 1–
6. doi: 10.1109/ICWITE57052.2022.10176235.
[52] M. Karimi and K. S. Viliyani, “Employee Turnover Analysis Using Machine Learning Algorithms”.
[53] J. Park, Y. Feng, and S.-P. Jeong, “Developing an advanced prediction model for new employee
turnover intention utilizing machine learning techniques,”Scientific Reports, vol. 14, no. 1, p. 1221,
2024.
AUTHORS
Haya Alqahtani is a student in the master’s program in Computer Information Systems at King Abdulaziz
University. Her research interests are machine learning, data science, and natural language processing.
Hana Almagrabi is an assistant professor in the Department of Information Systems at King Abdulaziz
University. Her research interests are in Text Mining and Natural Language Processing, Machine Learning,
and Data Science. Topics she has worked on include Information Extraction, Prediction Models, Text
Mining, and Content Analysis.
Amal Alharbi is an assistant professor in the Department of Information Systems at King Abdulaziz
University. Her research interests are in Natural Language Processing, Information Retrieval, Machine
Learning, and Data Science. Topics she has worked on include Information Extraction, Data Mining, and
Systematic Reviews.

EMPLOYEE ATTRITION PREDICTION USING MACHINE LEARNING MODELS: A REVIEW PAPER

More Related Content

What's hot (13)

Similar to EMPLOYEE ATTRITION PREDICTION USING MACHINE LEARNING MODELS: A REVIEW PAPER (20)

Recently uploaded (20)

EMPLOYEE ATTRITION PREDICTION USING MACHINE LEARNING MODELS: A REVIEW PAPER