Prediction of Hearing Problem Perception on the TrackYourTinnitus and TrackYourHearing Databases

Chronic diseases and conditions like tinnitus and hearing loss often have fluctuations when it comes to their perception by the people affected. Mobile self-reporting applications like TrackYourTinnitus and TrackYourHearing are helpful for keeping track of these variations and recording additional data. This thesis aims to explore connections between tinnitus and hearing loss perception on the one hand and this additional data on the other using machine learning techniques, namely Decision Tees and Support Vector Machines. In more detail, it tries to predict the self-reported disease perception using the answers to related questionnaires and data from the disease history of the user(s). These predictions are also performed on data from single users, revealing individual connections to the target, e.g. correlation with concentration or mood. They also achieved accuracy of up to 99%, albeit with some important caveats. The generalized prediction yielded acceptable results with 79% (tinnitus, binary decision) and 68% (60% non-weighted, hearing loss, 4 classes) accuracy. User-based prediction, however, could potentially be used to find out ways to help individuals by controlling the perception of the disease through associated variables (emotions).