Context-Aware Bias Detection and Mitigation in Machine Learning Models: A Study on Indian Socio-Economic Datasets

Social security services, loan approvals, job recommendation systems, and educational monitoring are only a few examples of the socioeconomic areas in which machine learning has emerged as a prominent area.Recent studies have shown that, contrary to popular belief, these applications can be used to maintain existing socioeconomic inequalities that exist within the training data.Most fairness-aware machine learning algorithms were developed in the West and fail to account for the various sociocultural factors that exist in India, such as caste, class, gender, geography, and rural-urban differences.To identify and mitigate biases in machine learning models trained on Indian socioeconomic factors, this paper proposes a context-aware approach.The study uses intersectional group analysis and basic fairness metrics to detect bias using a few data types and common machine learning classifiers.Experiments on a few mitigation strategies show that context-aware mitigation reduces bias significantly while preserving sufficient predictive accuracy.This study connects the theory of socio-cultural fairness to machine learning in the Indian context.