Predicting consumer loan default risk and systemic risk in financial institutions through machine learning techniques: a systematic literature review

Purpose This study aims to systematically review the application of machine learning (ML) techniques in predicting two major categories of financial risk: consumer loan default risk (CLDR) and systemic risk (SR) arising from institutional interconnectedness. The purpose is to identify the dominant models and methodological trends that define the current research landscape. Design/methodology/approach Using the preferred reporting items for systematic reviews and meta-analyses framework, this review initially retrieved 130 journal articles from the Scopus database, of which 50 met the inclusion criteria. The selected studies were comparatively analyzed to identify common prevailing ML techniques, data sets and evaluation metrics used for predicting CLDR and SR. Findings The review finds that models such as decision trees, support vector machines, naïve Bayes, random forest, LightGBM, XGBoost and artificial neural networks dominate CLDR prediction research. For SR, complex network embeddings methods, graph neural networks, recurrent neural networks (RNN), convolutional neural networks and network centrality measures are prevalent. Research limitations/implications The review also identifies contextual limitations and recommends incorporating hybrid models that combine deep learning with tree-based algorithms and explainable artificial intelligence interpretability for CLDR prediction. For SR, multivariate RNN-based models are recommended, and network-based analyses should be extended to include nonbank intermediaries, incorporating macroeconomic factors, policy environments and broader data sets in future research. Practical implications The findings offer valuable insights for implementing accurate, interpretable ML models to improve loan approval processes, credit risk monitoring and SR mitigation strategies. Social implications Enhanced financial risk prediction can strengthen economic stability, protect consumers from default and mitigate systemic crises, while interpretable ML models promote transparency and trust in automated financial decisions. Originality/value This study offers a comprehensive synthesis of ML approaches to financial risk prediction, emphasizing hybrid models, interpretability and broader systemic analysis. It serves as a guide for researchers and practitioners to develop more robust and context-aware predictive frameworks in financial institutions.