Resting state EEG based depression recognition research using voting strategy method

Depression has become a global mental illness that endangers people's health and imposes a serious burden on society. The diagnosis and treatment of depression in early stages are very important and can help more people to regain their health. In this work, 17 depressed patients and 17 normal subjects were recruited and participated in this study. 10 linear features and 8 nonlinear features from 16 electrodes in three frequency bands were extracted from the resting state EEG signals, including alpha, beta and theta bands. The feature search method GreedyStepwise (GSW) based on Correlation Features Selection (CFS) was used for dimension reduction. We used five classification algorithms to classify 17 depressed patients and 17 normal subjects, the highest classification accuracies obtained by the average strategy on the three bands were 68.92%±3.61%, 76.67%±1.72% and 76.76%±2.36% respectively. In order to improve the results, a voting strategy was employed in this paper, the highest classification accuracies obtained by using the voting strategy on the three bands were 77.65%±2.84%, 83.24%±2.42% and 81.47%±4.71% respectively, the results have been significantly improved compared to traditional methods. Our results show that the EEG-based classification methods combining vote strategy provide an accurate and convenient solution for depression recognition and could be used as an auxiliary tool for depression diagnoses.