Enhancing the Prediction of Employee Turnover With Knowledge Graphs and Explainable AI

Employee turnover poses a critical challenge that affects many organizations globally. Although advanced machine learning algorithms offer promising solutions for predicting turnover, their effectiveness in real-world scenarios is often limited because of their inability to fully utilize the relational structure within tabulated employee data. To address this gap, this study introduces a promising framework that converts traditional tabular employee data into a knowledge graph structure, harnessing the power of Graph Convolutional Networks (GCN) for more nuanced feature extraction. The proposed methodology extends beyond prediction and incorporates explainable artificial intelligence (XAI) techniques to unearth the pivotal factors influencing an employee’s decision to either remain with or depart from a particular organization. The empirical analysis was conducted using a comprehensive dataset from IBM that includes the records of 1,470 employees. We benchmarked the performance against five prevalent machine learning models and observed that our enhanced linear Support Vector Machine (L-SVM) model, combined with knowledge-graph-based features, achieved an impressive accuracy of 0.925. Moreover, the successful integration of XAI techniques for attribute evaluation sheds light on the significant impact of job environment, job satisfaction, and job involvement on turnover intentions. This study not only furthers the development of advanced predictive models for employee turnover but also provides organizations with actionable insights to strategically address and reduce turnover rates.