Employee Turnover Prediction Model Based on Feature Selection and Imbalanced Data Handling

Employee turnover prediction is vital for organizations aiming to retain talent and reduce operational costs. In this study, we utilized the Kaggle HR dataset to predict employee turnover through a comprehensive machine learning pipeline, including preprocessing, exploratory data analysis (EDA), and feature selection using correlation matrix analysis, Chi-Square tests, and Recursive Feature Elimination (RFE). To address class imbalance, we applied SMOTE and GAN-based oversampling techniques. We evaluated five models: SVM with CNN, Stacking Classifier, LightGBM, CatBoost, and a hybrid SVM + CNN model, assessing their performance using recall, F1-score, Cohen’s Kappa, ROC-AUC, PR-AUC, and confusion matrix. Results revealed that the hybrid SVM + CNN model, trained on GAN-balanced data and using RFE-selected features, achieved the best performance with an accuracy of 0.98, precision of 0.98, recall of 0.92, F1-score of 0.95, and Cohen’s Kappa of 0.94. The Stacking Classifier, LightGBM, and CatBoost also performed well, with results comparable to the hybrid model. The methodology was applied to the IBM HR dataset, where the same techniques improved minority class detection, demonstrating the robustness and generalizability of the framework across different datasets.

Keywords: Employee turnover prediction, human resource management, Kaggle HR dataset, machine learning, recursive feature elimination (RFE), SMOTE