From Weak Signal Detection to Crisis Anticipation: A Systematic Review of Machine Learning Models in Extreme Financial Risk Management

The increasing frequency of extreme financial events has revealed the limitations of traditional risk management approaches in capturing nonlinear dynamics and systemic disruptions. This study aims to examine the contribution of Machine Learning (ML) predictive models to the proactive management of extreme financial risks through a systematic literature review. Based on an analysis of 76 academic studies published between 2010 and 2024, the research identifies the main ML approaches used in risk prediction, including supervised, unsupervised, deep learning, and hybrid models. The findings show that ML techniques significantly improve the detection of weak signals, the anticipation of crises, and the modeling of complex risk dynamics compared with conventional statistical methods. However, several challenges remain, particularly data scarcity, class imbalance, concept drift, model interpretability, and organizational adoption constraints. The review further highlights the growing importance of Explainable Artificial Intelligence (XAI) and adaptive learning approaches in enhancing the reliability and transparency of predictive systems. The study contributes to the literature by proposing an integrated perspective linking predictive performance, explainability, and proactive risk management, while identifying key research directions for the development of more resilient and responsible financial risk management frameworks.