Enhancing Employee Attrition Prediction with Machine Learning and Feature Engineering

Employee attrition remains a critical challenge for organizations, leading to productivity loss and increased replacement costs. This study develops a machine learning framework to predict employee attrition, enhanced through domain-driven feature engineering. Using a publicly available HR dataset, we introduced four novel indicators—the Attrition Risk Score, Engagement Score, Income-to-Experience Ratio, and Commuting Stress Index that encapsulate key drivers of turnover such as engagement, compensation fairness, and work-life balance. A supervised learning pipeline was implemented and evaluated on five classification models. Among these, Logistic Regression achieved the strongest overall performance with 91.3% accuracy, an F1-score of 0.91, and a ROC-AUC of 0.963. The engineered features significantly improved predictive capacity and interpretability, with the Attrition Risk Score emerging as the most influential predictor. Results demonstrate that combining feature engineering with machine learning not only enhances model performance but also provides actionable insights for HR managers. Future work will extend this framework to broader datasets and real-world organizational settings.