RFRSF: Employee Turnover Prediction Based on Random Forests and Survival Analysis

In human resource management, employee turnover problem is heavily concerned by managers since the leave of key employees can bring great loss to the company. However, most existing researches are employee-centered, which ignored the historical events of turnover behaviors or the longitudinal data of job records. In this paper, from an event-centered perspective, we design a hybrid model based on survival analysis and machine learning, and propose a turnover prediction algorithm named RFRSF, which combines survival analysis for censored data processing and ensemble learning for turnover behavior prediction. In addition, we take strategies to handle employees with multiple turnover records so as to construct survival data with censored records. We compare RFRSF with several baseline methods on a real dataset crawled from one of the biggest online professional social platforms of China. The results show that the survival analysis model can significantly benefit the employee turnover prediction performance.