Systematic Review and Quantitative Evaluation of Advanced Machine Learning Frameworks for Credit Risk Assessment, Fraud Detection, And Dynamic Pricing in U.S. Financial Systems

Abstract

This study examined a key problem in U.S. financial services: advanced machine learning is used for credit risk assessment, fraud detection, and dynamic pricing, but decision value weakens when outputs are not trusted or audit ready across enterprise and cloud deployments. The purpose was to quantify how ML Framework Capability (MLC), the Model Confidence and Trustworthiness Index (MCTI), and Regulatory Readiness and Auditability (RRA) relate to domain effectiveness and overall decision-making effectiveness. Using a quantitative cross-sectional, case-based design, 168 valid responses were collected from cloud and enterprise financial services cases (banks 40.5%, fintech or platform lenders 26.2%, credit unions 14.3%, insurance or Insurtech 19.0%). Key variables were MLC, Credit Risk Effectiveness (CRE), Fraud Detection Effectiveness (FDE), Dynamic Pricing Effectiveness (DPE), Decision-Making Effectiveness (DME), MCTI, and RRA, all measured on 5-point Likert scales with strong reliability (Cronbach’s alpha .85 to .93). The analysis plan applied descriptive statistics, Pearson correlations, and multiple regression models predicting each domain outcome from MLC, MCTI, and RRA. Descriptives showed high perceived capability and outcomes (MLC M=4.11, SD=0.56; CRE M=4.02, SD=0.60; FDE M=4.18, SD=0.54; DPE M=3.86, SD=0.66; MCTI M=3.98, SD=0.57; RRA M=3.74, SD=0.63). MLC correlated with CRE (r=.62), FDE (r=.66), DPE (r=.53), and DME (r=.59), all p<.001. Regression results showed added value from trust and readiness: CRE R²=.49 with βMLC=.29 (p<.001), βMCTI=.41 (p<.001), βRRA=.17 (p=.006); FDE R²=.52 with βMLC=.34 (p<.001), βMCTI=.32 (p<.001), βRRA=.19 (p=.003); DPE R²=.40 with βMLC=.25 (p<.001), βMCTI=.35 (p<.001), βRRA=.14 (p=.021). Across domains, fraud detection showed the strongest perceived gains, while pricing was positive but more constrained, consistent with governance sensitivity. MCTI subdimensions indicated strongest audit traceability (M=4.17) and stability (M=4.01) but lower fairness confidence (M=3.83), highlighting a priority gap for deployment. RRA was driven most by audit trail logging (M=4.12) and documentation completeness (M=3.88). Implications are that institutions should operationalize trust-by-design through standardized explanations, fairness confidence checks, monitoring for drift, and auditable model version control to convert technical capability into defensible and consistent decisions.