Interpretable Gradient Boosted Modeling of Employee Attrition: A SHAP-Based Framework for HR Analytics

This study examines employee attrition using interpretable machine learning techniques, with a focus on enhancing strategic decision-making in human resource management. Three models—Logistic Regression, Random Forest, and Gradient Boosted Trees (GBT)—were evaluated, with GBT selected for its compatibility with SHAP (SHapley Additive exPlanations). SHAP was used to decompose the influence of variables such as Monthly Income, Over Time, and Job Satisfaction. The findings validate key HR theories, including Herzberg’s Two-Factor Theory and the Job Embeddedness framework, offering both predictive performance and theoretical alignment. The proposed model functions as a decision-support tool, providing actionable insights for HR professionals while contributing to the advancement of Management Technology through explainable AI.