Integrating Predictive Modeling, Algorithmic Fairness, and Explainable AI for Transparent Employee Performance Evaluation: An Empirical Illustration Using the IBM HR Analytics Dataset

Soniya Narayan

Department of Management Studies

Dayananda Sagar College of Engineering

Jayashree K

Assistant Professor

Department of Management Studies

Dayananda Sagar College of Engineering

Abstract
Artificial intelligence is changing the way employees are evaluated, but the challenge of concurrently high predictive power, demographic fairness, and decision transparency is a burning issue. The paper suggests and empirically proves a combination of ensemble predictive modeling (Random Forest), algorithmic bias, and SHAP-based explainable AI framework based on the organizational justice theory and strategic management views. As an example, with the publicly available IBM HR Analytics Employee Attrition & Performance dataset (1,470 records), Random Forest reached the maximum test accuracy (1.000), which is propelled largely by the job-related characteristic PercentSalaryHike. The analysis of fairness between gender groups presented a small difference, and the accuracy of the models was similar between employees both male and female. SHAP explanations affirmed that job-related predictors significantly dominate decisions whereas demographic attributes have very little effect on them- in line with employee perceptions of fairness. The results show that AI systems can provide high performance and transparency at a specific level when the relevant features of the job are prioritized. Theoretical implications of responsible AI in HRM, practical guidance to adopt AI in HR analytics, and implications to comply with regulatory requirements (e.g., EU AI Act) are addressed.

Keywords: AI-driven performance evaluation; algorithmic fairness; explainable AI; SHAP; HR analytics; organizational justice; bias mitigation; IBM HR dataset