Predicting employee attrition and explaining its determinants

An increased focus on utilizing data analytics to tackle human resource (HR) issues and make more informed and data-driven decisions is spreading in firms and public institutions. One of the major challenges faced by organizations is employee turnover, which can have negative impacts on productivity, performance, and overall corporate reputation. In light of these considerations, this study endeavors to predict employee attrition by deploying Machine Learning (ML) models on real-world data obtained from a prominent Italian financial corporation. Although the use of ML to predict attrition and investigate the main employers-employees features is documented in literature, what characterizes our study is the investigation of the crucial dimension of feature direction. Nonetheless, recognizing this directional aspect is pivotal for HR managers entrusted with making informed decisions. In our research, we employ the SHAP (SHapley Additive exPlanation) algorithm to not only identify feature contributions but also to assess their direction. Beyond mere algorithm implementation, our study interprets the outcomes within the specific context of HR decision-making. This comprehensive approach effectively highlights the inherent limitations of standalone algorithms, which may produce only partial results, capturing the importance of a feature, but missing its direction. Indeed, sometimes, while the feature is well known, its direction is somehow counterintuitive, thus requiring a deeper investigation and understanding. In a period like the present one, where the new production paradigms and the Covid-19 pandemic altered the consolidated labor market, new phenomena are emerging and only a profound understanding of the contextual novel dynamics can foster well-informed decision-making processes.