International Journal of Advanced Research in Computer and Communication Engineering

Abstract: As our society becomes increasingly autonomous and utilizes agentic systems, it is important to understand whether these systems subconsciously discriminate against certain populations based on characteristics such as employment status or education level. This research presents a machine learning framework to analyze loan approval decisions while ensuring algorithmic fairness across different demographics. Our study employs a systematic approach by combining multiple classification models with fairness analysis. To address class imbalance, we integrated into the pipeline. The framework evaluates Logistic Regression, Random Forest, Gradient Boosting, AdaBoost, and Support Vector Machines, utilizing fairness metrics such as True Positive Rates, False Positive Rates, and statistical uniformity across demographics. Results demonstrate that Gradient Boosting achieved the best performance, with CIBIL score emerging as the dominant predictive factor (86.8% feature importance), followed by loan term (9.7%) and loan amount (1.7%), while demographic characteristics showed minimal influence. Fairness analysis across education levels revealed approval rates of 34.81% for graduates versus 39.20% for non-graduates, though statistical testing (p=0.2086) indicated no significant bias. Similarly, employment status showed minimal disparate impact with only 0.56% difference in approval rates between self-employed and traditionally employed applicants (p=0.9221). The study contributes an analytical framework that shows how credit-relevant factors can drive lending decisions without introducing demographic bias; we achieved high accuracy (>97%) while maintaining fairness across protected groups.

Downloads: | DOI: 10.17148/IJARCCE.2025.1412120