Employee Turnover Prediction Using Machine Learning

Abstract

High employee turnover is a common problem that can affect organizational performance and growth. The ability to predict employee turnover would be an invaluable tool for any organization seeking to retain employees and predict their future behavior. This study employed machine learning (ML) algorithms to predict whether employees would leave a company. It presented a comparative performance combination of five ML algorithms and three Feature Selection techniques. In this experiment, the best predictors were identified using the SelectKBest, Recursive Feature Elimination (RFE) and Random Forest (RF) model. Different ML algorithms were trained, which included logistic regression, decision tree (DT), naïve Bayes, support vector machine (SVM) and AdaBoost with optimal hyperparameters. In the last phase of the experiment, the predictive models’ performance was evaluated using several critical metrics. The empirical results have demonstrated that two predictive models performed better: DT with SelectKBest and the SVM-polynomial kernel using RF.