Predicting employee turnover and reducing turnover costs using machine learning techniques

Abstract

This master thesis deals with the prediction and interpretation of voluntary employee turnover using machine learning and relates those findings to a cost reduction. In order to provide a theoret ical basis, different employee turnover costing frameworks, machine learning methods and feature importances from literature are evaluated. Based on these findings, insights into current turnover and retention costs for the company are obtained. Furthermore, historic employee turnover data is used to train and optimize six state-of-the-art machine learning models, which classify employ ees who are at risk of leaving. Ultimately, an optimal version of the best performing model is proposed and related to tangible insights. Individual classifications of this model are interpreted so that the insights can be used for global and individual retention strategies. A novel technique named SHAP is used to for this interpretation. Additionally, guidelines are provided on how to obtain these insights in novel situations when the model is used in practice. Lastly, the final model performance and interpretations are related to the turnover and retention costs, providing an indication of a possible cost reduction that can be achieved.