Explainable and Fairness-Aware Artificial Intelligence in Strategic Human Resource Management: A Dual-Module Machine-Learning Framework for Employee Retention and Promotion Decisions

AI has a lot of promise in the field of HRM, but relying on predictive outcomes isn't enough to make HR decisions. Two complementary open workforce datasets were used to create a machine learning dual-modules approach for employee attrition and promotion oriented to the governance. We used stratified benchmarking, nested cross-validation, paired out-of-fold classifiers, probability calibration and one-time untouched holdout testing to evaluate the conventional and advanced classifiers. Precision-recall discrimination, calibration, threshold performance, explainability, demographic fairness, proxy detection, review-capacity evaluation, and demographic fairness were all assessed using a model. The best holdout PR-AUC, ROC-AUC, precision and Matthews correlation coefficient for attrition were obtained with a sigmoid-calibrated linear support vector machine, with values of 0.6025, 0.7944, 0.7000 and 0.4968, respectively. The CatBoost model with an isotonic calibration was chosen for promotion and has the following metrics: PR-AUC of 0.5503, ROC-AUC of 0.8193, precision of 0.8734 and Matthews correlation coefficient of 0.5516. Explainability analysis found stable workforce predictors and subgroup and proxy audits showed that removing sensitive attributes from the workforce did not remove any demographic information or disparity risks. Positive net benefit and large lift over random review suggested that there is potential decision utility under constrained review capacity. The results endorse calibrated machine learning as an augmentative decision-support system, which is subject to global thresholds, fairness, contextual evaluation, and compulsory human supervision and not automated hiring choices.