Auto Machine Learning for Early Warning Crisis Detection

In this work, we investigate the application of Automated Machine Learning (Auto ML) techniques to Early Warning Systems (EWSs) for financial crises at the country level. We explore the end-to-end process of applying supervised learning with Auto ML to financial crisis detection, including data preparation, the treatment of imbalanced datasets without relying on synthetic data, and the construction of representative training and testing sets. This study shows how Auto ML can partially automate the selection and tuning of machine learning pipelines (namely, sequences of computational steps, such as preprocessing and classification), thereby reducing the manual effort required while achieving competitive performance. The results indicate that Auto ML provides a valuable starting point for developing financial crisis EWSs, although further manual refinement and ad hoc adjustments of the identified pipelines remain necessary.