FAIRLENS: An Explainable, Fairness-Aware Platform for Bias Detection in Workforce Evaluation

Abstract

Workforce evaluation is where organizational bias does its quietest damage: subjective peer reviews and managerial discretion systematically under-reward some groups and overlook “silent performers,” and the decisions are rarely explained. This paper presents FAIRLENS, a fairness-aware and explainable platform that audits and corrects bias in employee evaluation rather than merely automating it. On an HR dataset with realistic injected bias—subjective peer-review and collaboration signals skewed against an unprivileged group—a standard performance-prediction model (AUC 0.90) is shown to inherit that bias, producing a disparate impact of 0.30, far below the 0.8 legal threshold, and an aggregate Bias Risk Index of 0.40. FAIRLENS detects this with four fairness metrics and then mitigates it by group-aware post-processing, driving disparate impact to 1.0 and cutting the Bias Risk Index by $70 \%(0.40 \rightarrow 0.12)$ at a modest accuracy cost measured against the biased label. The platform makes every score explainable (permutation feature attributions identify the biased peerreview signal as the dominant driver), defines a Contribution Intelligence Score that recovers true merit better than the biased managerial score r = 0.69 vs 0.56), predicts burnout from workload signals (AUC 0.96), and uses collaboration-graph centrality to surface silent performers. With defined algorithms, fairness evidence, explainability, and an ethics-and-privacy framework, FAIRLENS becomes a verifiable contribution to responsible workforce analytics.