MACHINE LEARNING-BASED PREDICTIVE ANALYTICS FOR CREDIT RISK ASSESSMENT: A COMPREHENSIVE REVIEW

Abstract

Credit risk assessment is one of the most consequential analytical tasks in modern finance, underpinning lending decisions, regulatory capital calculations, and the overall stability of banking systems. For decades, financial institutions relied on statistical scoring methods such as logistic regression and discriminant analysis, but the growing volume, velocity, and variety of financial data have accelerated a shift toward machine learning (ML) and deep learning (DL) approaches. This paper presents a comprehensive review of the literature on ML-based predictive analytics for credit risk assessment, synthesizing evidence from twenty-three peer-reviewed studies, review articles, and book chapters published between 2019 and 2025. The review traces the evolution of credit risk modeling from classical statistical techniques to ensemble learning, deep neural architectures, and hybrid frameworks; examines their application across retail credit scoring, corporate financial distress prediction, credit card fraud detection, rural and microfinance lending, banking supervision, and sector-specific contexts such as shipping finance; and consolidates commonly used datasets and evaluation metrics. The synthesis indicates that ensemble methods (particularly gradient-boosted trees such as XGBoost) and deep learning architectures generally outperform traditional statistical baselines, especially for nonlinear and high-dimensional data, while hybrid and ensemble combinations tend to yield the most consistent gains across benchmark datasets. Persistent challenges include class imbalance, limited model interpretability, data privacy constraints, regulatory compliance, and inadequate data infrastructure in emerging markets. The paper concludes by outlining research directions centered on explainable AI, federated and privacy-preserving learning, and the development of standardized, representative benchmark datasets for credit risk research.