DEVELOPING FAIRNESS METRICS FOR FACIAL RECOGNITION MODELS APPLIED TO CENTRAL ASIAN ETHNIC GROUPS

Facial recognition systems are increasingly used across Central Asia in security, governance, and digital identity applications. However, most existing models are trained on datasets that do not adequately represent regional ethnic groups, resulting in biased outcomes. This study focuses on developing fairness metrics tailored to Central Asian populations, including Uzbeks, Kazakhs, Kyrgyz, Tajiks, and Turkmens. The research critiques the limitations of current evaluation practices that rely on broad racial categories and overlook subgroup disparities. A methodology is proposed involving balanced dataset construction, ethnicity-specific categorization, subgroup-wise accuracy assessment, and analysis of error rate asymmetry. Anticipated outcomes include the identification of performance gaps, cross-ethnic misclassification patterns, and fairness deviations not captured by standard metrics. The findings are expected to support the development of localized standards for ethical AI deployment in the region. By introducing context-aware fairness evaluation, this work advances efforts toward equitable and inclusive facial recognition technologies