Explainable Machine Learning and Graph Neural Network Approaches for Predicting Employee Attrition

Abstract

Employee attrition constitutes a remarkable challenge for most organizations, usually leading to loss. To survive in the modern, highly competitive business world, the organization should be able to predict attritions for strategic planning and retention of critical employees. Artificial intelligence's (AI) potential to predict employee attrition has been identified as a powerful tool we can harness to improve employee retention. Whereas traditional machine learning models achieve high accuracy (e.g., XGBoost: 95.55% accuracy, 98.44% recall), they fall short due to the black box nature. This work leverages explainable graph neural networks (GNNs) to predict employee departure and identify critical factors influencing their decisions. The GNN-based model offers explainability, which is crucial for understanding why an employee is likely to leave. We leverage the GNN model's ability to label the deep-rooted structure of employee data, where connections between colleagues can hold valuable insights. Our model provides interpretable predictions by incorporating explainable AI techniques, highlighting the most influential factors contributing to employee turnover. Our work contributes to AI interpretability discourses, and our results allow HR professionals to develop targeted interventions and retention strategies based on the specific reasons identified by the model. Although the GNN achieved a lower overall accuracy than some traditional models, our focus on explainability offers valuable insights for HR decision-making. This research paves the way for building trustworthy AI and Responsible intelligent systems for employee retention in organizations.