Browsing by Author "Raziyeva S."

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    Mitigating Bias in AI-Based Loan Approval Systems through Fairness-Centric Techniques
    (SDU University, 2025) Raziyeva S.
    As artificial intelligence (AI) becomes increasingly embedded in high-stakes decision-making systems, ensuring fairness in algorithmic outcomes has emerged as a critical concern. This thesis investigates bias and fairness in AI-based credit scoring systems, with a particular focus on gender disparities. Using the German Credit Dataset as a case study, the research evaluates the performance and fairness of several supervised machine learning models, including Logistic Regression, Decision Tree, Random Forest, XGBoost, Support Vector Machine, and Neural Network. The study applies fairness metrics such as Statistical Parity Difference (SPD) and Disparate Impact (DI) to assess group-level inequalities in predicted loan approval outcomes. Results reveal a consistent trade-off between model accuracy and fairness, where high-performing models like Random Forest and XGBoost demonstrate notable biases against female applicants. Even interpretable models, such as Logistic Regression, exhibit fairness issues due to historical and structural biases embedded in the training data. To address these challenges, the thesis highlights the importance of incorporating fairness-aware strategies across the machine learning pipeline, including data pre-processing, fairness evaluation, and potential post-processing mitigation. The use of tools like AIF360 and stratified sampling further strengthens the analysis. This research contributes to the growing discourse on responsible AI by demonstrating that achieving fairness is not merely a technical goal but a sociotechnical imperative. It calls for an interdisciplinary approach that combines ethical reasoning, regulatory compliance, and algorithmic transparency to ensure equitable access to financial services. The findings advocate for the development of AI systems that are not only accurate but also accountable and inclusive.