AttritionGuard: Combining Explainable AI with Deep Neural Networks for Enhanced Employee Attrition Prediction

Abstract

"Employee attrition, the phenomenon of employees leaving an organization, poses a critical challenge for businesses worldwide, resulting in increased costs, loss of talent, and disruption in operations. Identifying and addressing the underlying factors contributing to employee attrition is essential for maintaining organizational stability and productivity. This research addressed the challenge of predicting employee attrition by developing a Deep Neural Network (DNN) based model. The DNN architecture consisted of three hidden layers with a decreasing number of neurons (15, 10, and 5) to balance the model complexity with the ability to learn intricate patterns in employee data and prevent overfitting. To enhance the model interpretability, eXplainable Artificial Intelligence (XAI) techniques, specifically SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), were integrated. The DNN model achieved a high level of performance with an accuracy of 95.71%. Additionally, precision, recall, specificity, F1-score, and ROC AUC score all exceeded 0.95, demonstrating the model's effectiveness in identifying employees at risk of leaving. Furthermore, both global and local explanations generated by the XAI techniques aligned with the established knowledge in Human Resource Management, providing valuable insights into the factors influencing employee attrition."