Discrimination of psychotic symptoms from controls through data mining methods based on emotional principle components

Abstract

In this study, different data mining techniques has been used for classification of healthy controls and patients diagnosed by First Episode Psychosis with respect to complexity of frequency band activities (Delta, Theta, Alpha, Beta, Gamma)in multi channel EEG measurements mediated by emotional, static and visual stimuli including affective pictures from IAPS. Degree of local EEG complexity has been correlated by largeness of the dominant principle component in each EEG sub-band. The best classification performances are provided by Rotation Forest, Simple Logistic and Artificial Neural Networks when the components from occipito-parietal and posterio-temporal locations (P3, P4, O1, O2, T5 and T6) are considered as features in Gamma with respect to neutral emotional state.