REPRESENTATION LEARNING FOR IMPROVED INTERPRETABILITY AND CLASSIFICATION ACCURACY OF CLINICAL FACTORS FROM EEG