Studying the Potential of Multi-Target Classification to Characterize Combinations of Classes with Skewed Distribution
The identification of subpopulations with particular characteristics with respect to a disease is important for personalized diagnostics and therapy design. But what if the manifestation of the disease is not described by one target variable but of many? Multi-target classification algorithms are the straightforward choice in this context and have been successfully applied in different application scenarios. However, most investigations do not focus on the effects of a skewed class distribution, where the prevalence of one of the multi-target combinations is more rare than the others. Moreover, in personalized medicine, it is not only essential to separate subpopulations but also to characterize them in a human understandable way. In this study, we analyze the potential of multi-target classification for the identification and characterization of subpopulations, that exhibit higher prevalence for a rare combination of targets. We report on the results of our approach for the analysis of tinnitus screening data with respect to two target variables, tinnitus loudness and handicap.