Predicting currency and financial crises has garnered a lot of attention and research, with notable developments and a variety of methodological approaches. This review article summarizes current research from 2019 to 2024 with an emphasis on the kinds of datasets, timeframes, and models used. The examination displays a wide range of data, from the 1970s to 2022, with the majority of studies
... [Show full abstract] depending on data collected after the 1990s because of its increased dependability and availability. Many models have been used, such as Markov switching models, artificial neural networks (ANN), signal approaches, deep neural decision trees (DNDTs), and traditional econometric models like logit and probit. The results emphasize that there isn't a single model that is always better; rather, they emphasize the significance of choosing models based on context and the advantages that hybrid or ensemble approaches may have. Our review highlights that to improve prediction accuracy, a variety of datasets and models must be used. Because currency crises are inherently complex, a multifaceted strategy that makes use of both conventional econometric and contemporary machine learning techniques is required. To better capture the complex dynamics of economic indicators, future research should investigate higher frequency data and keep improving hybrid methodologies. This thorough analysis contributes to the current discussion on currency and financial crisis forecasting by offering insightful analysis and directing future research paths.