BALANCING ACCURACY AND INTERPRETABILITY IN CREDIT RATING MODELS: A COMPARATIVE ANALYSIS OF STATISTICAL AND MACHINE LEARNING METHODS IN NIGERIA

Abstract

Credit rating remains one of the most critical mechanisms in financial markets, serving as a benchmark for investment decisions, borrowing costs, and regulatory compliance. This study investigates the predictive capacity of machine learning and traditional statistical methods for corporate credit rating in Nigeria, an emerging economy where reliable credit risk assessment remains central to financial stability and capital allocation. Adopting an explanatory research design, the study utilizes firm-level panel data from a purposive sample of 89 publicly listed companies operating across manufacturing, financial, and service sectors over the period 2015–2023. Using data from 89 companies across three major sectors, the study evaluates the performance of six models, including logistic regression, support vector machines, random forest, XGBoost, decision trees, and k-nearest neighbours, based on classification accuracy, sensitivity, specificity, precision, and Matthews correlation coefficient. Empirical results reveal that while advanced machine learning algorithms such as random forest and XGBoost perform competitively, logistic regression demonstrates consistent interpretability and regulatory suitability, particularly when applied to capital adequacy and profitability indicators. Sensitivity analyses and post-hoc Tukey HSD tests suggest that differences across models are not always statistically significant, underscoring the role of data quality and multidimensional indicators in determining predictive success. The study concludes that hybrid credit-rating frameworks that integrate interpretable statistical models with machine learning techniques offer the most practical balance between transparency and predictive power. It recommends that regulators and financial institutions in Nigeria prioritize data enrichment strategies, encourage explainable modeling approaches, and incorporate alternative data sources to enhance corporate credit assessment and support financial stability.