Explainable Data Analytics in Financial Decision Systems: Enhancing Transparency in Big Data–Driven Credit Risk and Loan Approval Models

Abstract

This study examined explainable data analytics in big data–driven credit risk and loan approval decision systems using a quantitative, model-comparison research design. A cross-sectional dataset consisting of 1,248 loan application records was analyzed within a retail lending context to evaluate how explainability-related attributes functioned alongside conventional predictive performance indicators. The empirical analysis incorporated applicant demographic and financial profile variables engineered behavioral features, and multiple explainability constructs, including explanation stability, sparsity, and governance-related traceability measures. The observed loan approval rate was 63.8%, while the default or delinquency rate within the performance window was 11.9%, indicating a moderately imbalanced but analytically viable outcome distribution. Descriptive results showed that predicted risk scores were concentrated in lower-to-moderate ranges, with a mean score of 0.284 and limited out-of-time drift, reflected by a score stability index mean of 0.087. Explainability summaries indicated concentrated attribution structures, with the top three explanatory drivers accounting for an average of 64.8% of total contribution magnitude and an explanation sparsity mean of 4.2 features per decision. Reliability testing demonstrated acceptable to excellent internal consistency across composite constructs, with Cronbach’s alpha values ranging from 0.81 to 0.92. Regression analysis revealed that explainability stability (β = 0.842, p < 0.001) and governance traceability (β = 0.311, p < 0.001) were statistically significant predictors of loan approval outcomes after controlling for applicant risk and financial variables, improving model fit from a pseudo-R² of 0.241 to 0.293. A secondary regression model showed that higher explanation stability and sparsity were associated with lower error concentration, with the explainability stability index exhibiting a negative relationship with misclassification clustering (β = −0.118, p < 0.001). Overall, the findings demonstrated that explainability, when operationalized through measurable stability, sparsity, and governance indicators, functioned as a statistically observable dimension of model quality and decision consistency within credit risk and loan approval systems.