Employee Turnover Prediction Research of Human Resource Management on Machine Learning Algorithms and Big Data Analysis

Abstract

This study introduces a new tool for predicting employee turnover using machine learning (ML) and big data. This method integrates LightGBM and XGBoost (both weighted 1, with predictions summed) to enhance accuracy and stability. To improve model interpretability, the SHAPT model is used to identify key factors affecting turnover, such as salary, position, and tenure. Experimental results show the integrated model outperforms standalone LightGBM and XGBoost: accuracy is 1.5% higher, F1 value is 0.02 higher, and AUC reaches 0.9504. These validate the model; SHAP analysis also provides actionable HR management insights, enabling early identification and response to potential employee departures. The research offers practical tools for HR decision-making. Future work will incorporate additional socio-economic variables and dynamic data to further improve prediction performance.