IV Course on Machine Learning and Central Banking

The Directorate of Financial Market Infrastructures of CEMLA, together with the Deutsche Bundesbank, organized the IV Course on Machine Learning and Central Banking. The course was held virtually from April 19 to 21, 2023. It was attended by 89 representatives of 23 CEMLA institutions and partners, representing 21 countries in Latin America and the Caribbean, 3 European countries and 1 Asian country. During the course, the basic components of Machine Learning (ML) were presented and some methods were analyzed in detail, establishing connections between them and conventional statistical methods.

During the first day of the course, introductory topics needed to train an ML model were covered. In particular, the differences between training, validation and test data were discussed. Similarly, the concept of cross-validation was addressed and all its relevant details discussed. Next, ways to evaluate an ML model were studied, among which the confusion matrix stands out, as well as measures of precision, sensitivity, F1-score and the PR curve.

On the second day of the event, some tree-based methods, such as decision trees (CART) and conditional inference trees, were discussed. Later, some criteria such as the GINI index and entropy were explained, which are used so that models can make distinctions between the data. During the last session of the day some more advanced models of trees such as Bagging, and Boosting were covered. While the former is based on sampling techniques, the latter is based on the premise "can several weak algorithms create a more robust algorithm?".

During the last day of the course, the so-called Random Forests were studied, as well as Gradient Boosting, which improve some qualities of the previous methods. However, it was stressed that care must be taken in its implementation, since, depending on the qualities of the training data, these may have a lower performance than the introductory algorithms covered in the previous days.  During the last session of the course, real problems of some central banks were discussed. Participants who presented their problems included members of the Central Bank of Nicaragua and the Central Bank of the Philippines.