Incremental Semi-Supervised Fuzzy C-Means for Bipolar Disorder Episode Prediction

Bipolar disorder is a chronic mental illness characterized with changing episodes (depression, mania, mixed state, euthymia). In the recent years, smartphone becomes an increasingly important tool in the early prediction of a starting episode. Usually, the state of the art research applies supervised learning methods and first of all, limits the dataset only to those days that have valid labels (from the psychiatric assessment), secondly, ignores the time structure of data. We pursue an alternative approach and apply incremental semi-supervised fuzzy learning without the need to limit the dataset only to labeled data. As observed, it is able to adapt the model as new data arrive. Preliminary results show that the algorithm is able to detect some of healthy episodes (euthymia) and disease episodes even when only 25% of labels are available.