Using machine learning algorithms to classify data has always been a complex task as the algorithms have many parameters (so called hyperparameters). Optimizing these has been a task for experts for a long time now. In the last few years there have been attempts to automatically optimize the hyperparameters. The limitation of these approaches is that they assume the concrete processing pipeline and the ranges of the hyperparameters for each algorithm used to be given.
In this work a new approach is taken, to utilize the knowledge of domain- and algorithm-experts to automatically determine the optimal pipeline and optimize the hyperparameters for these. Therefore care must be taken during generation and optimization of the different pipelines, because the number of possible pipelines might be very large.
