Identifying and Mitigating Data Bias in AI-Generated Test Cases

As artificial intelligence (AI) increasingly influences financial decision-making, ensuring fairness in automated sys- tems has become crucial. AI models for loan approvals often replicate historical biases, disadvantaging groups such as women, rural applicants, and individuals with limited credit history. This study presents a structured approach to detect and mitigate bias in AI-generated test cases by combining the Synthetic Minor- ity Oversampling Technique (SMOTE) with fairness evaluation methods. The framework includes exploratory data analysis, baseline model training, synthetic edge-case generation, and bias mitigation through balanced sampling. Experiments on a real-world loan dataset show a 25% reduction in gender-based disparities and 18% reduction across education levels while maintaining 92% predictive accuracy. The method is applicable to credit scoring, fraud detection, and other high-stakes financial systems to promote equitable AI-driven decision-making.