A data-driven decision-support framework for prioritizing retention interventions among high attrition-risk employee segments

Predicting employee attrition risk provides a useful signal for human resource decision-making. However, this signal alone does not determine which employee groups should be targeted first for retention interventions. This study develops a data-driven decision-support framework for prioritizing retention interventions among employees at risk of attrition. The framework uses the IBM HR Analytics Employee Attrition and Performance dataset and integrates calibrated machine learning, SHAP-based explanatory analysis, Entropy-CRITIC weighting, and PROMETHEE II ranking. Machine-learning models were first used to estimate employee-level attrition probabilities. These probabilities were then aggregated into segments defined by combinations of Department, JobRole, and OverTime, from which the high-risk segment pool was identified for multi-criteria analysis. Eight high-risk segments were detected. The segment ranked first by PROMETHEE II was not the segment with the highest attrition risk, indicating that intervention priority and attrition risk represent distinct decision outputs. Sensitivity analyses further showed that the segment-ranking structure remained stable under alternative weight-shift and threshold scenarios. The proposed framework is therefore not intended to identify individual employees at risk; rather, it supports transparent, auditable, and ethically compliant segment-level retention decisions.