Data mining for prediction of human

International Journal of Data Mining & Knowledge Management Process (IJDKP) Vol.5, No.2, March 2015
DOI : 10.5121/ijdkp.2015.5205 53
DATA MINING FOR PREDICTION OF HUMAN
PERFORMANCE CAPABILITY IN THE
SOFTWARE-INDUSTRY
Gaurav Singh Thakur1
, Anubhav Gupta2
and Sangita Gupta3
1
Cisco Systems, Bangalore, India
2
Common Floor Technologies, Bangalore, India
3
Jain University, Bangalore, India
ABSTRACT
The recruitment of new personnel is one of the most essential business processes which affect the quality of
human capital within any company. It is highly essential for the companies to ensure the recruitment of
right talent to maintain a competitive edge over the others in the market. However IT companies often face
a problem while recruiting new people for their ongoing projects due to lack of a proper framework that
defines a criteria for the selection process. In this paper we aim to develop a framework that would allow
any project manager to take the right decision for selecting new talent by correlating performance
parameters with the other domain-specific attributes of the candidates. Also, another important motivation
behind this project is to check the validity of the selection procedure often followed by various big
companies in both public and private sectors which focus only on academic scores, GPA/grades of students
from colleges and other academic backgrounds. We test if such a decision will produce optimal results in
the industry or is there a need for change that offers a more holistic approach to recruitment of new talent
in the software companies. The scope of this work extends beyond the IT domain and a similar procedure
can be adopted to develop a recruitment framework in other fields as well. Data-mining techniques provide
useful information from the historical projects depending on which the hiring-manager can make decisions
for recruiting high-quality workforce. This study aims to bridge this hiatus by developing a data-mining
framework based on an ensemble-learning technique to refocus on the criteria for personnel selection. The
results from this research clearly demonstrated that there is a need to refocus on the selection-criteria for
quality objectives.
1. INTRODUCTION
Software engineering is a discipline of science which deals with systematic development of
software. It has various methods, process and tools to achieve high quality products. However
software-quality still isn’t up-to the mark. Practitioners are realizing that people are the biggest
risk if not performing well. Software quality can be achieved by cumulative quality standards of
process and people over all generic modules in development [10]. It is utmost important to look
into the people component deeply in the people, process, product triad. Data mining is the
process of extracting useful knowledge from data [8]. It utilizes a combination of a knowledge
base, sophisticated analytical skills and domain specific knowledge to uncover many hidden
trends and patterns. These patterns and relations can be extracted by using various data-mining
algorithms. Today, the human aspects of software engineering have become one of the critical

54
concerns in IT companies to achieve their business goals. Industries be it in any sector, are now
paying attention to selecting the right talent who can perform consistently well throughout all
generic framework activities and execute the process properly. Software quality is greatly
dependent on the people and process quality during the development and testing phase. Hence, in
this paper we use data mining algorithms to exploit the patterns in the historical data and predict
the performance based on project-personnel attributes and thereby enhance the quality of process
and assure project success. Data Mining is the next big revolutionary field that is redefining the
industry, be it in terms of technology or research. Here we use mathematical procedures like
function-approximation techniques to solve prediction or estimation problems and extract useful
trends and patterns from past data to facilitate ourselves to take right decisions with the aim to
produce near-optimal results. Classification, allows us to identify association rules.
Categorization uses induction algorithm rules to handle categorical outcomes, such as good,
average and poor as in this study.
In this paper we address the issue of developing an ideal selection framework for recruiting the
right talent which brings us to the basic question of what criteria to follow for the selection
procedure. Our aim is to understand the relationship between the various project-personal
attributes of the candidates and their professional-performance parameter as rated by their
managers/supervisors in the industry. Our aim is to identify the variables which have the
maximum predictive power in estimating the performance capabilities of new recruits. Though
there have been many previous studies in this domain, there have been certain issues that still
need to be addressed. We try to build upon them and use this research to build a better and more
robust model that can be not be applied to different scenarios but also work well on different data
sets having varied properties with minimal or no changes. There are a wide range of algorithms in
each of these categories, many of which are implemented on WEKA and R [7]. These tools are
platforms with GUI and command-line implementations respectively, with a number of machine-
learning algorithms for data mining tasks, with a variety of options for regression, classification,
data pre-processing, association rules, clustering and visualization.
The paper is organized as follows: Section 2 provides the related work and background of data
mining Algorithms. Section 3 presents the research methodology to derive at a conclusion.
Section 4 discusses about the implementation details while Section 5 depicts the obtained result.
Section 6 discusses the summary of this paper and its future scope.
2. RELATED WORK
With increasing complexity of software in the industry and their ever growing demands in
multidisciplinary projects, there has been a continuous progress in research works that target the
areas of effective project management and Data mining has recently proven itself as one of the
most established techniques in this area. Data mining methodologies are developed for several
applications including various aspects of software development and we plan to employ this power
of algorithms to develop a selection framework. There have been a plethora of studies which
incorporate the tools of machine-learning for developing a framework for Prediction of Human-
Capability.
The authors in [13] have studied the importance of different variables that come into play during
the selection of students - like academic scores , Programming Skills, Domain Knowledge
Assessment, Reasoning skills, Mental Ability and Mathematical skills and many more using

55
Decision Trees ID3, CART and C4.5. As it turns out, there are many features that must be tested
along with academic scores to confirm the quality of new personnel. However, the model used in
this project has significant scope for improvement and we base our project on the hypothesis
similar to this research by Sangita Gupta et. al. to further improve upon the results obtained using
a different approach that involves Ensemble-learning technique – Random Forest, a bagging
based approach to accuracy boosting. Focusing on other works, authors in [4] surveyed different
machine learning algorithms for defect prediction in software. Authors of [5] have used data
mining technique clustering and done a comparative analysis of performance of Density-Based
Spatial Clustering of Applications with Noise (DBSCAN) and Neuro-Fuzzy System for
predicting the level of severity of faults in Java-based object oriented software. Data mining
results in decision through methods & not through assumptions. Authors in [2] have worked on
the improvement of employee selection in semiconductor industry by developing a model, using
data mining techniques. The specified attributes involved age, gender, marital-status, experience,
education, major-subjects and school-tires as potential factors that might affect the performance.
As an outcome of their study, it was found that employee-performance is greatly affected by
educational-degree, the school-tire, and the work experience. The authors in [3] researched on
multiple factors that affect the job performance of employees. They reviewed previous works
which study the effect of experience, salary and training, working-conditions and job satisfaction
on the performance-parameters. Data-mining thus supports various techniques including
Statistical-Analysis, Decision-Trees, Genetic-Algorithm, Bayes-classification, visualization
techniques, etc. for analysis and prediction. It further facilitates association, clustering and
classification [1]. This research-study involves applying data in WEKA and R tool and derives a
classification model for selection criteria. The algorithm we use in this project is Random-Forest,
a bagging based accuracy boosting technique which improves upon the result published in [13]
which uses Decision Trees namely ID3, C4.5 and CART.
3. RESEARCH METHODOLOGY
The research methodology followed during this study involved the following steps as stated
below:
Figure 1 : Framework for Building recruitment Process

56
• Hypothesis: Project personnel with similar skills set and capabilities will perform
similarly.Project-personal information about the employees is collected from projects
which use similar technology and programming language and work on similar platforms.
Thus personnel under comparison here have similar capabilities and skill-sets.
• Data Collection: Data was collected by using various techniques such as Form-Filling,
Brainstorming, obtaining performance information about employees from the Project-
leads and managers.
• Data Preparation: A basic preliminary research concluded with an almost same set of
attributes that must be considered under this problem-statement to obtain a correlation
with performance parameters as in [13]. The list of attributes noted are as follows:
I. PS - Programming Skills: The programming and coding skills of employees were
tested using standing coding problems with multiple test cases. A complete score
was awarded for each problem only if all the test cases were executed
successfully. Employees were tagged as good for scoring above an overall value
above 75%, average if they scored between 50% and 75% and the rest were
tagged as poor.
II. RAS – Reasoning and Analytical Skills: Similar to programming skills, RAS
values were collected through an internal assessment which involved analytical
and logical-reasoning questions.
III. DSK - Domain Specific Knowledge: Categorized and normalized similar to PS
and RAS. Obtained from a series of internal assessments, DSK refers to both
theoretical and field-based knowledge about the domain in which the employees
have been working.
IV. TE - Time efficiency: Marked as a simple "Good" or "Bad" option by the project
manager/leader of an employee which shows if the employee is considered
efficient in completing his/her work within the expected time-period.
V. GPA – Grade Point Average: This is the grade scored by an employee during
his/her graduation/ post-graduation courses. It is categorized and normalized to
scale the GPAs on the same level based on universities.
VI. CS - Communication Skills: This score for each of the employees was obtained
from the project-leaders, their team mates and the respective HR-manager of the
employee as well.
VII. P – Performance Parameter: This is our dependent variable for the problem
statement which must be correlated to the independent attributes available from
the preliminary study of relevant features. The value was acquired by
brainstorming with the Team-leads and Managers assigning an overall
performance score to the employees in terms of good, average and poor.

57
This preliminary-study of relevant attributes is followed by the selection of a suitable machine-
learning model that must be adopted for the data-mining process. As we have seen, a number of
studies have already been done in this domain but the issue has been lack of high-accuracy. Such
accuracy issues can be addressed very well if accuracy-boosting techniques are applied. Also
given the fact that Knowledge-based Decision Trees have been known to perform decently in
such cases, we employ the bagging based Ensemble-learning model Random forest in this
scenario to further enhance the performance of old models.
Underlying Learning Model - Random Forest
In a classification problem, we have a training sample of n observations on a class variable Y that
takes values 1, 2, ... k, and p predictor variables, X1,..., Xp. Our goal is to find a model for
predicting the values of Y from new X values. In theory, the solution is simply a partition of the
X space into k disjoint sets, A1, A2,..., Ak, such that the predicted value of Y is j if X belongs to
Aj , for j = 1, 2,..., k. If the X variables take ordered values, two classical solutions are linear
discriminant analysis1 and nearest neighbor classification. These methods yield sets Aj with
piecewise linear and nonlinear, respectively, boundaries that are not easy to interpret if p is large.
Classification tree methods yield rectangular sets Aj by recursively partitioning the data set one X
variable at a time. This makes the sets easier to interpret. For example, Figure 1 gives an example
wherein there are three classes and two X variables. The left panel plots the data points and
partitions and the right panel shows the corresponding decision tree structure. A key advantage of
the tree structure is its applicability to any number of variables, whereas the plot on its left is
limited to at most two.
Pseudocode for tree construction by exhaustive search is as follows:
• Start at the root node.
• For each X, find the set S that minimizes the sum of the node impurities in the two child
nodes and choose the split {X∗ ∈ S∗} that gives the minimum overall X and S.
• If a stopping criterion is reached, exit. Otherwise, apply step 2 to each child node in turn.
Since, this is a Regression problem, we use CART for producing individual regression trees and
then build upon this model by using the Random Forest technique to achieve better results.
CART uses a generalization of the binomial variance called the Gini index. A regression tree is
similar to a classification tree, except that the Y variable takes ordered values and a regression
model is fitted to each node to give the predicted values of Y. This yields piecewise constant
models. Although they are simple to interpret, the prediction accuracy of these models often lags
behind that of models with more smoothness. It can be computationally impracticable, however,
to extend this approach to piecewise linear models, because two linear models (one for each child
node) must be fitted for every candidate split.
These trees may not provide very high accuracies, since they have very high variance values.
Randomization based ensemble methods, prove to be a good solution to this flaw. Random-Forest
consists of a collection or ensemble of simple tree predictors, each of which outputs a response
when presented with a set of predictor values just as the input vector X. For classification-based
problems, this response can be of the forms - class membership or associations, a set of

58
independent predictor values with one of the categories present in the dependent variable. Each
tree is created from its own separate bootstrapped sample training set. The Bootstrap Sampling
Method samples the given training tuples uniformly with replacement i.e. each time a tuple is
selected, it is equally likely to be selected again and rendered to the training set. As the number of
simple learning models within an ensemble technique increases, the overall variance of the
output-value from the actual-value theoretically decreases by 1/(number of individual models).
However this decrease in variance after a threshold doesn’t yield significant improvement and
that allows us to decide the number trees we want to create for this random forest technique. It’s
important to remember that ensemble learning techniques are computationally expensive and
hence choosing an optimal value for the number of individual simple predictors within the
ensemble technique is a critical task. Individual Decision-trees usually suffer from high-variance,
which makes them uncompetitive in terms of accuracy. A highly efficient and simple way to
address this issue is to adopt the context of randomization and use them in ensemble-methods.
The Mean-Decrease-in-Accuracy of a variable is evaluated during the calculation-phase of out-
of-bag error. As the fall in accuracy of the random-forest increases due to the addition of a single-
variable, the more important the particular variable under test is considered and hence variables
with a large value for Mean-Decrease-in-Accuracy or Gini are considered as more important for
data classification. The Mean-Decrease-in-Gini coefficient is a measure of how each particular
variable supplements to the homogeneity of the nodes and terminal-leaves in the resulting
Random-forest. Every time one particular variable splits a given node, the Gini coefficient for the
children are calculated and compared to that of the original parent node. If the same variable
causes multiple splits more than once, then the final difference in the Gini value of the topmost
parent node and the bottom-most children nodes is taken as the Mean-decrease-in-Gini value. The
pseudo code for generation of a random forest is as follows:
i. Draw number_of_trees bootstrap samples from the original data.
ii. Grow a tree for each bootstrap data set. At each node of the tree, randomly select mtry
variables for splitting. Grow the tree so that each terminal node has no fewer than
nodesize cases.
iii. Aggregate information from the ntree trees for new data prediction such as majority
voting for classification.
iv. Compute an out-of-bag (OOB) error rate by using the data not in the bootstrap sample.
Using these values, a final graph for Variable Importance is plotted, where this graph represents
each variable on the vertical y-axis, and their importance-values on the horizontal x-axis. They
are ordered in the manner of top-to-bottom as maximum-to-minimum-importance. To measure
the accuracy of the classifier, we made use of Sensitivity and Specificity parameters. The
following are the meaning of the variables used in the subsequent equations.
• True positive = correctly identified
• False positive = incorrectly identified
• True negative = correctly rejected

59
• False negative = incorrectly rejected
1. True-Positive Rate or Sensitivity is the fraction of training samples predicted correctly by
model.
2. False-Positive Rate or Specificity i.e. the fraction of training samples predicted incorrectly by
model.
where FPR represents the False positive Rate and lower this value, the better the model is.
3. Area under ROC curve (Receiver Operating Characteristics): is obtained by plotting TPR
against FPR. The area under the plotted graph gives a good measure of the accuracy of the
classifier. The area can be as high as 1 Sq. unit (maximum accuracy) and as low as 0
(minimum accuracy). Since Area-Under-ROC-curve takes into consideration, both FPR and
TPR values, this measure is preferred over the parameters to compare accuracies between
models.
4. IMPLEMENTATION DETAILS
Based on the study conducted, data obtained was consolidated and summarized in a tabular form.
The algorithm used here is Random Forest, implemented using WEKA and R under "Test
options". 10 - Fold Cross validation was applied to supplement the out-of-bag calibration
mechanism of Random-Forests. The number of individual predictor-trees was set to 500 in R as
no significant reduction in variance was observed beyond this value. Trees were allowed to grow
completely without any pre or post pruning. The package used in R for implementing Random
Forest is the “randomForest” package which is compatible with versions 4.6 and above. This
package is available for download at the official R support website.
For creating bootstrap samples, we used the technique of 632 Bootstrapping, which means that in
any bootstrap sample generated, approximately 63.2% of the Dataset will be unique, and the rest
would be placed with replacement and duplication. Studies have shown that this bootstrapping
technique produces near-optimal results. The plot for Variable Importance has been obtained
using R tool and the results obtained are discussed below. The model generated cannot be
visualized graphically due to the large number of trees generated, each created from a separate
bootstrap sample and each producing its own results. As stated earlier, the final output from the
Random Forest model is the average of the results obtained from each of the predictor trees in the
forest.
5. RESULTS AND DISCUSSION
Table 1 shows the results obtained by Random forest. Table 2 shows the Variable Importance
values for each of the attributes in terms of Mean-Decrease in Gini and Mean-Decrease in
Accuracy.

60
Table 1 : Accuracy Measures for samples classified by output class
CLASS TP RATE FP RATE Area under ROC curve
Good 0.934 0.037 0.977
Average 0.846 0.037 0.983
Poor 0.929 0.077 0.992
Firstly we look at the average area under the ROC curve and as we can see, this area is about
0.984 (average). The area here is close to 1 and for ROC curves; an area 1 refers to the highest
possible accuracy of 100%. Hence, we can see that the Random-Forest Model used is highly
accurate and strong conclusions can be drawn from these results. We also found through a
comparative study that the model outperformed knowledge-based decision trees and Linear
Regression techniques when applied to the same data.
We now take a look at the Variable importance plot using Mean-Decrease in Gini and Mean-
Decrease in Accuracy values.
Table 2 : Variable Importance measures for each Attribute
DSK RAS PS CS TE GPA
Mean Decrease Accuracy 35 10 9 0 0 0
Mean Decrease Gini 14 5 4 2 1 2
Figure 2 Plot for Variable Importance of different attributes
As we see in these plots, the variables like Domain-Specific knowledge, Reasoning and
Analytical skills and Programming Skills are the most important attributes to be considered
during recruitment of new personnel, as they have the maximum contribution towards the
homogeneity of nodes and classification of data. Another important result obtained in this
scenario is the Mean Decrease in Accuracy value for GPA which comes out to be 0. Such low
scores for these values stand as a scientific base to challenge the usual recruitment procedure

61
where maximum importance is given to GPA of candidates. What we see here is that, based on
the grades and academic scores of students from universities, one cannot predict their
performance in the industry. Hence GPA alone is not a very clear reflection of the candidate’s
capabilities as far as the software industry is concerned. It is needed that the companies separately
test the other relevant attributes of the students in order to make a better decision. The software
development process involves various intricate steps and complex stages where many other
factors and abilities of an individual come into play. Referring to the GPA alone will not yield
optimal results and this is the reason, why there has been a changing trend in the recruitment
procedure today. Recruiting teams are looking for candidates with a complete package in terms of
overall personality, analytical thinking abilities and good inter-personal skills apart from good
grades. The importance given to personal interviews lays emphasis on the fact that only after
getting a true idea about the candidate’s knowledge in the particular domain through a one-on-
one interaction, companies take a final decision regarding the selection procedure.
Hence, using the results from the variable importance plot, we create ideal knowledge based
procedure consisting of rules which can be used as the selection criteria in appropriate scenarios.
Also, we use the Mean-Decrease-in-Gini Parameter for laying these rules to obtain maximum
homogeneity. The rules-based procedure is as follows:
Figure 3 Suggested procedure to be followed While Recruiting
As we see, during the selection process companies must initially clarify the values for each of the
variable-parameters that are acceptable to them. For example some companies may accept
students with average programming skills but others may only want those who have great
programming skills whatever may be the scores for other attributes. Once that’s done, they must
start classifying students based on the features in the order of their variable importance values.
Another important aspect here is to remember that in numerous scenarios, there may be a very
large feature set. In such a case, it is not possible for the companies to consider all of them.
Hence, to modify the selection tree, a tree pruning mechanism is developed as follows:

62
1. Set the threshold limit that must be used to prune to the tree, say ‘P’ percentage of
pruning is needed.
2. Evaluate the percentage importance values using the following formula. If for variable Xi
the Mean-Decrease-in-Gini score is given as Imp(Xi) then evaluate Percentage Imp(Xi) as
3. Now select the variables with minimum ܲ݁‫ݐ݊݁ܿݎ‬‫݌݉ܫ‬(ܺ݅) scores and sum them until this
total greater than or equal to ‘P’.
4. All those variables Xi with minimum ܲ݁‫ݐ݊݁ܿݎ‬‫݌݉ܫ‬(ܺ݅) values that were added to the sum ≥
P should not be used during the construction of the ideal-tree.
This procedure allows us to cut down on the least important variables to be considered during the
selection-procedure using the pre-pruning technique. Such a method would be helpful in
scenarios when the number of variables is very high and it’s not practical for the companies to
look at every attribute. In case of a tie, it totally depends on the discretion of the hiring manager
to exclude or include those features from the ideal-selection tree. Though this model provides
high accuracy and allows pruning techniques to adapt to real-time scenarios, there is still scope
for further improvement using cost optimization techniques like Gradient Boosting and many
more, which would yield better results when dealing with higher dimensional data and inter-
dependent variables. In order to develop a selection-framework for other domains, a similar
procedure can be followed after incorporating minimal changes to suit them appropriately.
6. CONCLUSION
The primary target of data-mining is to produce near optimal results using the information
extracted from patterns and trends hidden in historical data. This brings us to the essential
question of choosing the best suitable model that can be applied to any given problem statement.
In this scenario, the models of Decision trees have already been tested on a similar dataset with
similar attributes and properties in [13]. However, this model suffers from the problem of
overfitting and we try to overcome this issue by applying the Random-Forest technique. We also
see that even when most studies provide a prioritization mechanism for selection procedure, they
lack a quantifying measure to represent their importance in the framework.
The Random Forest method is a bagging based accuracy boosting technique, which creates
multiple trees out of the bootstrap samples generated from a dataset, hence forming a forest. Here
the output from each tree is considered to calculate an overall mean or an average result for the
random forest. Since the calibration of the model is done using the out-of-bag samples, the model
does not suffer from over-fitting issues which facilitates a superior performance in most of the
scenarios when compared to Decision-trees. Apart from these enhancements in accuracy,
Random forests also provide supplemental information like variable importance measures, etc.

63
which adds to its value. One of these measures is the Mean Decrease in Gini index and Mean
Decrease in Accuracy parameters which gives a holistic view about the contribution of each of
the attributes to the final output. We utilize this facility provided by Random-Forest to assign an
order of priorities to the features that must be considered during the recruitment of new-
personnel.
We clearly see from the results that the GPA/grades of new recruits clearly aren’t among the most
important selection attributes on which a hiring-decision can be based. Other attributes like
Programming Skills, Domain Specific Knowledge and Analytical Skills must be separately tested
as they have a significant prediction power in estimating the performance of a person in the
software industry. However, only obtaining a number of important parameters using a simple
model do not suffice this need as the degree of accuracy associated with the model is equally
important. This critical need for an accurate model would be clearly visible when developing a
framework for other domains and working on high-dimensional data and this where Accuracy-
boosting techniques like random forests come into play. These algorithms are not only known for
their robustness but also perform well in cases of high-dimensional data and sparse data
sets.Data-Mining algorithms extract important patterns and have helped in identification of
project-members who have a greater probability to perform well. This research not only
empowers the managers to refocus on Human-Capability criteria to enhance the development-
process of any software-project but also address the issue where due to lack of analytical-methods
in human-aspects, IT companies could not select the right talent in the software process and
hence failed to achieve the desired-objectives in terms of both, quality and quantity in the time
and cost-constraints. The scope of this study can be further extended to various domains and
appropriate changes need to be incorporated when developing a selection-framework for each one
of them. Such a study would require a preliminary research about finding the relevant features
that can be directly correlated to the performance of any given employee in that particular
domain. Once this is done, a similar model can be used to develop a feature-ranking order and
enlist the most important attributes to be considered to make a hiring-decision using such a
framework.
REFERENCES
[1] Jiawei Han and Micheline Kamber, “Data Mining Concepts and Techniques, 2/e”, Morgan Kaufmann
Publishers, An imprint of Elsevier, 2010.
[2] C.F. Chien and L.F. Chen, "Data mining to improve personnel selection and enhance human capital:
A case study in high-technology industry", Expert Systems and Applications, vol. 34, 2008, pp. 280-
290.
[3] HamidahJantan , "Human Talent Prediction in HRM using C4.5 Classification Algorithm", (IJCSE)
International Journal on Computer Science and Engineering Vol. 02, No. 08, 2010, pp. 2526-2534.
[4] Suma.V, Pushpavathi T.P, and Ramaswamy. V, “An Approach to Predict Software Project Success
by Data Mining Clustering", International Conference on Data Mining and Computer Engineering
(ICDMCE'2012), pp. 185-190.
[5] P. Singh, "Comparing the effectiveness of machine learning algorithms for defect prediction",
International Journal of Information Technology and Knowledge Management, 2009, pp. 481-483.
[6] J. R. Quinlan, "Introduction of decision tree", Journal of Machine learning, 1986, pp. 81-106.
[7] Witten I. Frank E., and Hall M., “Data Mining: Practical Machine Learning Tools and Techniques”,
3rd Edition, Morgan Kaufmann Publishers, 2011.
[8] A. Kusiak, J. A. Kern, K. H. Kernstine, and B. Tseng, “Autonomous decision-making: A data mining
approach,” IEEE Trans. Inform. Technol. Biomedicine, vol. 4, no. 4, pp. 274–284, Aug. 2000.

64
[9] A. S. Chang, &Leu, S.S., "Data mining model for identifying project profitability variables,"
International Journal of Project Management, vol. 24, pp. 199-206, 2006.
[10] T. R. Gopalakrishnan Nair, V. Suma, Pranesh Kumar Tiwari," Analysis of Test Efficiency during
Software Development Process",2nd Annual International Conference on Software Engineering and
Applications (SEA 2011)
[11] J. R. Quinlan, “C4.5: Programs for Machine Learning”, Morgan Kaufmann Publishers, Inc, 1992.
[12] OdedMaimon, LiorRokach, “The Data Mining and Knowledge Discovery Handbook”, Springer
publication,2005.
[13] Sangita Gupta , Suma V, " Empirical study on selection of team members for software project- A data
mining approach" , International Journal of Computer Science and Informatics, ISSN (PRINT): 2231
–5292, Vol 3, no 2, 2013,pp 97-102.Sangita Gupta, Suma V: Prediction of Human Performance
Capability during Software Development using Classification. ICT and Critical Infrastructure:
Proceedings of the 48th Annual Convention of Computer Society of India- Vol II, Advances in
Intelligent Systems and Computing Volume 249, 2014, pp 475-483
AUTHORS
Gaurav Singh Thakur
Gaurav Singh Thakur has completed his B.Tech in 2014 in Information Technology
from National Institute of Technology Karnataka, Surathkal and is currently working
as a Software Engineer at Cisco Systems, Inc. Bangalore. His technical areas of
interest include Machine learning, Networking & Security, Application Development
and Algorithms.
Anubhav Gupta
Anubhav Gupta has pursued his Bachelors, at National Institute of Technology
Karnataka, Surathkal in the Field of Information Technology (IT) and graduated in
2014. His technical areas of interest include Machine learning, Information Security,
Web Development and Algorithms. Currently, he is working as Software Developer
Engineer at Commonfloor (MaxHeap Technologies).
Sangita Gupta
Sangita Gupta is a research Scholar of Jain University Bangalore. She has done her
B.Sc, M.Sc and M.Phil in computer science and guide for B.tech, M.Phil students

Data mining for prediction of human

More Related Content

What's hot (19)

Similar to Data mining for prediction of human (20)

Recently uploaded (20)

Data mining for prediction of human