Explainable Employee Churn Prediction using SHAP-based Interpretability Analysis

The problem of employee turnover is critical to organizations both in terms of operations and finance. In turn, such a requirement of human resource (HR) analysis as valid and reliable predictive models is of paramount importance in the modern context of contemporary human resource (HR) analysis. Even though machine learning models have demonstrated excellent predictive power regarding employee turnover, a lack of explainability renders them less desirable in real-life situations that involve the human resources department. The current paper presents a comparative study of the application of explainable machine learning models to predict employee turnover utilizing demographic, occupation-related, compensation variables, and employee satisfaction data in the human resource dataset of an organization. This paper discusses some predictive models, which are both explainable and inherently explainable, in the context of predictive accuracy and explainability. These diverse frameworks ensure that the various stakeholders within the HR sector are able to not only point out employees at risk of leaving their workplace but also the causes of employee attrition. Besides, the paper reflects the use of Responsible AI Guidelines through equity evaluation, bias testing in sensitive variables, and transparent results in the research. The performance of the models through experiments justifies their competencies in that the models can perform equally well with more desirable explanations for employee retention strategies.