Hyperparameter optimization of two-branch neural networks in multi-target prediction

Dimitrios Iliadis, Marcel Wever, Bernard De Baets, Willem Waegeman

Abstract: Highlights•Multi-target prediction (MTP) in an umbrella term that covers many different sub-areas.•DeepMTP was already proposed as a single approach that is compatible with the most MTP problem settings to date.•DeepMTP already automates many different aspects of the typical machine learning pipeline but not the hyperparameter optimization step.•We benchmark different hyperparameter methods using popular datasets from multiple MTP problem settings.•Various comparisons motivate why we select Hyperband as the best approach for our framework.