The integration of artificial intelligence (AI) in the financial sector has seen a rapid increase over the past few years, offering new possibilities to streamline processes while ensuring profitability for lending institutions. With its data-driven capability, predicting the creditworthiness of applicants has demonstrated strong predictive performance, particularly for thin-file clients. Despite these advances, growing concerns regarding AI’s fairness, explainability, and regulatory accountability have increasingly limited its adoption in high-stakes credit decision-making. This paper presents a synthesis derived from a systematic literature review (SLR) of 43 peer-reviewed studies published between 2020 and 2025, focusing on AI-based credit scoring and addressing at least one of the performance, fairness, or explainability dimensions. Eligible studies were limited to peer-reviewed journal and conference articles (2020–2025) retrieved from IEEE Xplore, Scopus, Web of Science, and ScienceDirect (last searched: 30 September), examining AI-driven credit scoring in consumer or lending decision contexts. Guided by the Relevance, Rigor, Reproducibility, and Quality (3Rs&Q) appraisal framework, the review analyzes how existing approaches navigate the interplay among performance, fairness, and explainability under regulatory and human oversight considerations. The findings indicate that these dimensions are predominantly addressed in isolation, with limited attention to their joint treatment in regulated deployment settings. By consolidating empirical and conceptual evidence, this review provides actionable guidance for designing and deploying credit scoring models in practice.

Feature selection is pivotal in enhancing the efficiency of credit scoring predictions, where misclassifications are critical because they can result in financial losses for lenders and exclusion of eligible borrowers. While traditional feature selection methods can improve accuracy and class separation, they often struggle to maintain consistent performance aligned with institutional preferences across datasets of varying size and imbalance. This study introduces a FastTree-Guided Genetic Algorithm (FT-GA) that combines gradient-boosted learning with evolutionary optimization to prioritize class separability and minimize false-risk exposure. In contrast to traditional approaches, FT-GA provides fine-grained search guidance by acknowledging that false positives and false negatives carry disproportionate consequences in high-stakes lending contexts. By embedding domain-specific weighting into its fitness function, FT-GA favors separability over raw accuracy, reflecting practical risk sensitivity in real credit decision settings. Experimental results show that FT-GA achieved similar or higher AUC values ranging from 76% to 92% while reducing the average feature set by 21% when compared with the strongest baseline techniques. It also demonstrated strong performance on small to moderately imbalanced datasets and more resilience on highly imbalanced ones. These findings indicate that FT-GA offers a risk-aware enhancement to automated credit assessment workflows, supporting lower operational risk for financial institutions while showing potential applicability to other high-stakes domains.

As regulatory requirements increasingly shape automated lending decisions, fairness remains a critical challenge in high-stakes domains, particularly credit scoring. Although artificial intelligence models can achieve strong predictive performance, they may also reproduce biased outcomes that reduce financial inclusion or transfer harm to overlooked protected groups. Existing fairness interventions commonly operate at a single stage of the decision-making pipeline, despite bias often propagating across representational and decision layers. This study proposes iCert-Fair, a two-layer framework for technical fairness assessment and harm recovery in credit scoring. The first layer adopts a fairness-through-explainability paradigm, using SHAP-based explanations to identify direct and proxy dependence on protected attributes and guide structural dataset repair, while the second layer applies targeted threshold-policy adjustments to recover residual harm while preserving decision utility. Experiments on the German and Taiwanese credit datasets show that fairness gains are model- and dataset-specific and may be collective, concentrated, transferred, or recovered unevenly across protected attributes. The direct comparison with representative pre-processing, in-processing, and post-processing methods revealed that baseline methods targeting one protected attribute at a time frequently transferred residual harm to other monitored attributes. In contrast, the fairness-focused recommendations generated by iCert-Fair achieved larger collective fairness improvements across all considered protected attributes while avoiding residual harm. These gains were obtained while preserving predictive utility on the German dataset and with utility degradation remaining below 5% across the evaluated performance metrics on the Taiwanese dataset, alongside consistently lower false-negative risk. The empirical findings support the use of complementary structural and policy-level interventions and demonstrate the importance of jointly evaluating aggregate disparity, worst-case attribute-level harm, cross-attribute transfer, and predictive utility.