An End-to-End Framework for Predicting Employee Attrition with Explainable AI

Employee attrition is a persistent challenge for organizations, generating significant costs and impacting productivity. While machine learning can predict employee turnover, common approaches struggle with imbalanced data and lack interpretability for actionable insights. This paper presents a complete, end-toend framework for attrition prediction that is both accurate and transparent. Using the IBM HR Analytics dataset, the methodology features advanced preprocessing, engineered attributes, and class balancing via the Synthetic Minority Oversampling Technique (SMOTE). A Random Forest classifier is employed, withmodel evaluation based on Recall, F1-Score, Precision, and AUC-ROC — metrics aligned with the business goal of accurately identifying at-risk employees. To ensure decision transparency, Explainable AI tools SHAP and LIME are integrated, providing both global and local interpretability. Results reveal that the proposed model identifies 80% of actual attrition cases while highlighting key drivers such as overtime workload, compensation levels, and job satisfaction. This interpretability supports targeted retention strategies instead of generic interventions. The study demonstrates that combining machine learning withXAI techniques enables both reliable prediction and actionable insights. Future work includes integrating career progression timelines and deploying real-time attrition monitoring dashboards.