As part of the M.Sc. program at the Eindhoven University of Technology (TU/e), I worked 3 months for Achmea. The goal of this internship was to develop and apply an AutoML solution to help the company to come up with better ML models, particularly for fraud detection. I worked under the supervision of Leon Vink, Cyril Cleven, Senna van Iersel, and Peter Diks.

Achmea's campus at Tilburg.


Our solution consisted of a meta-learning module and an evolutionary algorithm. For the meta-learning part, we used a database of meta-features that characterize public datasets, building a meta-space with which one can find the most similar dataset to a new one, and take its model as a possible solution. Later, with the evolutionary algorithm, the candidate models can be fine-tuned and a winner is selected.

We made the tool open source. You can can check it out in the website of the project and the Git repository.