Artificial Intelligence and Machine Learning for Economic Crisis Prediction and Early Warning: A Systematic Review

Economic crises inflict severe damage on employment, output, and social welfare, making their early detection a priority for policymakers and financial institutions. Traditional early warning systems, built on logistic regression and signal-extraction methods, have shown limited out-of-sample accuracy and an inability to capture the nonlinear dynamics characteristic of modern financial systems. Over the past decade, machine learning (ML) and artificial intelligence (AI) methods have emerged as promising alternatives. This paper presents a systematic review of the literature on AI and ML applications in economic and financial crisis prediction, covering 47 studies published between 2008 and 2025. I categorize the methods into five families: tree-based ensembles, neural networks and deep learning, support vector machines, natural language processing approaches, and hybrid architectures. The review reveals that ensemble methods—particularly random forests and gradient boosting—consistently outperform logistic regression, with Bluwstein et al. (2023) reporting AUROC values of 0.870 for extremely randomized trees versus 0.822 for logistic regression across 17 countries from 1870 to 2016. Credit growth and yield curve slope are identified as the most robust predictors. Deep learning shows promise for temporal dependencies but faces data scarcity challenges. NLP-based approaches using central bank communications represent a rapidly growing frontier. We identify key challenges including event rarity, interpretability demands, and concept drift, and conclude with a research agenda for the field.