Fairness in predictive analytics: integrating bias detection, mitigation, and explainability in machine learning models

Abstract

This thesis presents a critical analysis of machine learning algorithms within the realm of educational predictive analytics, with a particular emphasis on detecting and mitigating socio-economic biases. The research employs an analytical framework comprising bias detection techniques to identify inherent biases in algorithms or datasets, bias mitigation models to adjust these elements and reduce socio-economic disparities, and explainability methods to elucidate the decision-making mechanisms of the algorithms.