Employee Retention Prediction: A Data Science Project by Devangi Shukla
- 2. Employee Retention Prediction Descriptive Objective: This project focuses on predicting employee attrition using machine learning to help companies anticipate when employees are likely to leave. By analysing historical HR data, including factors like job role, tenure, and demographics, the goal is to predict turnover and take proactive retention measures. This approach falls under HR Analytics, which leverages data to improve workforce management. Importance : By predicting attrition, organizations can implement retention strategies like improving career development or enhancing workplace conditions. This also allows businesses to plan future hiring needs, reducing the costs of turnover and improving workforce stability. In summary, this solution helps companies make informed, data-driven decisions to retain talent, reduce churn, and improve long-term success
- 3. The Challenges of Employee Turnover Financial Impact Turnover costs businesses significant resources in recruitment, training, and lost productivity. Team Morale Losing valuable employees disrupts team dynamics and can negatively impact morale. Loss of Expertise The departure of experienced employees can lead to a loss of institutional knowledge and expertise.
- 4. Factors Influencing Employee Retention 1 Compensation and Benefits Competitive salaries and benefits packages are essential for attracting and retaining talent. 2 Work-Life Balance Employees value flexible work arrangements and opportunities to prioritize their well-being. 3 Career Growth Opportunities Clear paths for advancement, training, and development motivate employees to stay.
- 5. INFORMATION ON THE DATASET company_size: Size of the company where the enrollee works or worked. city: City where the enrollee is located. city_development_index: Indicator of the development level of city. enrollee_id: Unique identifier for each enrollee. major_discipline:Academic field or discipline of the enrollee’s major. relevent_experience:Whether the enrollee has relevant work experience. experience: Number of years of work experience. education_level: Highest level of education attained by the enrollee. gender: Gender of the enrollee. training_hours:Total number of hours spent on training. company_type:Type or category of the company.
- 6. DATA UNDERSTANDING AND INSIGHTS EDA- EXPLORATORY DATA ANALYSIS HANDING MISSINGVALUES DATA ENCODING AND OUTLIER DETECTION MODEL BUILDING – LR, XG, RF CONCLUSION LIST OF CONTENTS
- 7. FOLLOWING LIBRARIES HAVE BEEN USED Description of these libraries are as follows:- * Pandas for Dataframe operations * Numpy for Numeric operations * Matplotlib and Seaborn are Data Visualisation libraries * Scikit-Learn for all the Machine learning algorithms and
- 8. Descriptive Statistics Column Mean Std Dev Min 25th Percentile 50th Percentile (Median) 75th Percentile Max city_develo pment_inde x 0.775 0.075 0.624 0.748 0.776 0.834 0.920 experience 6.4 6.5 <1 <1 5 >20 >20 training_ho urs 45.2 30.4 8 47 52 83 83 target 0.6 0.5 0 0 1 1 1 Descriptive Statistics for Numerical Columns:
- 9. Descriptive Statistics Descriptive Statistics for Categorical Columns:
- 10. OBSERVATIONS FORTHE DESCRIPTIVE STATISTICS city_development_index Training hours Experience Attrition (Target) Gender shows a moderate variation in city development (ranging from 0.624 to 0.920). heavily skewed towards higher values (many employees have more than 5 years of experience, with a few having <1 year or >20 years). range widely, with some employees having very high training hours (up to 83 hours). has a fairly balanced distribution between 0 and 1, suggesting that both employees who stay and leave are relatively equally represented. predominantly male (80% of the entries), while education level is mostly Graduate (80%).
- 11. DATA VISUALISATION The dataset is primarily concentrated in cities with higher development indices (e.g., 0.920), while cities with lower indices (e.g., 0.625) have minimal representation. The dataset shows a significant gender imbalance, with the majority of entries being male (17729), followed by a smaller number of females (1238) and a few entries with missing or unspecified gender (191).
- 12. The dataset indicates that most employees (13792) have relevant experience, while a smaller group (5366) lacks relevant experience. The majority of employees have no enrolment in university programs (14203), followed by those enrolled in full-time courses (3757), with a smaller number in part-time courses (1198). DATA VISUALISATION
- 13. DATA VISUALISATION The dataset shows that most employees have a graduate education level (11598), followed by those with master's degrees (4361), and fewer with doctoral or other higher education levels. The dataset reveals that the majority of employees come from STEM disciplines (14492), followed by those with a Business Degree (2813), and smaller groups with other disciplines like Arts and Humanities. The dataset shows the largest groups of employees work in companies with 50-99 employees (3083), followed by 100-500 employees (2571), with smaller representation in larger companies (10000+ and 5000-9999).
- 14. HANDLING MISSINGVALUES THESE ARE THE MISSINGVALUES IN EACH COLUMN IN PERCENTAGES : THAT SHOWS: Major data is missing for the following variables : Gender (22.5%) Major_Discipline (15%) Company_Size (31%) Company_Type (32%)
- 15. We cleaned the data and removed the missing values through the below code where we defined the *cleanNaN* function with parameter (dfa) That resulted to the cleaned data:
- 16. OUTLIER DETECTION AND Z- SCORE Performing z-test to remove outliers in Train Dataset Performing z-test to remove outliers in Test Dataset Outlier detection using the Z-score involves calculating the number of standard deviations a data point is away from the mean; points with a Z-score above a threshold (e.g., |3|) are considered outliers. These outliers can be removed or capped to reduce their impact on modeling.
- 17. Normalization of Data The resultant data after sampling, needs to be normalized between certain range of values, so that the model wont be biased towards the high values of different variables. The data has been normalized to the values between 0 & 1, independently of the statistical distribution they follow. Train andTest data splitting *The data has been split to test data, training data & the model is trained with the training data. * Both the dependent & independent variables perform with higher precision when the unknown test data is fed are split to test & training data. * It is done, in order for the model to to it. TRAIN ANDTEST DATA SPITTING
- 18. BUILDING MACHINE LEARNING MODELS LOGISTIC REGRESSION Logistic Regression is a Machine Learning classification algorithm that is used to predict the probability of a categorical dependent variable. Logistic Regression is classification algorithm that is not as sophisticated as the ensemble methods. Hence, it provides us with a good benchmark.
- 19. BUILDING MACHINE LEARNING MODELS DECISION TREE CLASSIFIER Decision tree classifier is a machine learning algorithm used for both classification and regression tasks, that predicts value of a target variable by learning simple decision rules inferred from the input features. * Decision trees are structured as a hierarchical tree-like structure, where each internal node represents a feature or attribute, and each branch represents a decision rule based on that attribute. The leaf nodes represent the final predicted outcome or class label.
- 20. BUILDING MACHINE LEARNING MODELS RANDOM FOREST CLASSIFIER Random Forest is a machine learning method that is capable of solving both regression and classification. It is a brand of Ensemble learning, as it relies on an ensemble of decision trees. It aggregates Classification (or Regression) Trees. * Random Forest fits a number of decision tree classifiers on various sub-samples of the dataset and use averaging to improve the predictive accuracy and control over-fitting. Random Forest can handle a large number of features, and is helpful for estimating which of your variables are important in the underlying data being modeled.
- 21. CONFUSION MATRIX, ROC CURVE AND AUC SCORE The model performs well overall, with an accuracy of 84.2%, but it struggles with recall (43.4%) for detecting attrition (employees who leave).This indicates that the model is better at predicting employees who stay (True Negatives) than those who leave. Precision (56.8%) and F1-Score (49.3%) suggest that while the model is better at predicting employees who stayed, its predictions for employees who left are less reliable, as evidenced by a moderate number of False Negatives (539) and False Positives (314). AUC (Area Under the Curve): The AUC score is 0.7802, which indicates that the model has good discriminative ability. An AUC of 0.5 would indicate random guessing, while 1 would indicate perfect classification.
- 22. GRID AND RANDOM SEARCHING FOR FINE TUNING OF HYPER PARAMETERS Grid search works by creating a grid of all possible combinations of hyperparameter values specified by the user. It then trains and evaluates model using each combination of hyperparameters and selects the one that yields the best performance based on a predefined evaluation metric, such as accuracy, precision, or F1 score It systematically explores all possible combinations of hyperparameters, ensuring that the best combination is found within the specified search space. However, this exhaustive search can be computationally expensive. Random Search is a hyperparameter tuning technique where random combinations of hyperparameters are sampled and evaluated to find the best-performing model. It's an alternative to Grid Search that can be more efficient, especially when the hyperparameter space is large. Random search can explore a wide range of hyperparameters quickly, while grid search can be computationally expensive if the grid is large.
- 23. GRID SEARCH FOR RANDOM FOREST CLASSIFIER We are taking the results of the RF as it has proven to be the best performer overall
- 24. RANDOM SEARCH FOR RANDOM FOREST CLASSIFIER We are taking the results of the RF as it has proven to be the best performer overall
- 25. Employee turnover is driven by 3 key factors: Lack of Career Advancement Employees may leave if they feel there are limited opportunities for growth or advancement within the company. Inadequate Compensation and benefits Poor Work Culture and Environment When pay and benefits don't align with industry standards or employee expectations, workers may seek better offers elsewhere. A toxic work environment, lack of recognition, or ineffective leadership can lead to dissatisfaction, causing employees to leave in search of a more supportive and rewarding workplace.
- 26. Conclusion: The dataset provides insights into factors influencing employee attrition, such as experience, education, and tenure. Understanding these factors helps develop targeted retention strategies. In model performance, Random Forest (RF3) leads with the highest F1 score (0.73) and accuracy (0.78), offering a strong balance between precision, recall, and overall performance. XGBoost (XGB3) follows closely, with an F1 score of 0.72 and similar accuracy. Decision Tree Regressor (DTR) excels in precision for class 0 but struggles with class 1, while Logistic Regression performs poorly across metrics. Overall, Random Forest is the best model for predicting employee attrition.