Depression Diagnosis Based on KNN Algorithm and EEG Signals

This work aims to diagnose depression and isolate healthy people from depressed patients based on EEG brain signals via the k-nearest neighbor algorithm (KNN) and using 10-fold cross-validation. Five regular frequency bands (Gamma, Beta, Alpha, Theta, and Delta) were utilized from the signals. Band power and median band frequency were extracted by Welch’s periodogram method as features. After classification, the highest accuracy gained by using frequency features in the left hemisphere was from the Alpha and Beta waves which resulted in 100% output (p<0.05), and as for the right hemisphere highest accuracy was for the Gamma, alpha, and Beta oscillators, which also resulted in 100% (P<0.05). the lowest accuracy was assigned to the Delta band. In general, combining the two hemispheres boosted the accuracy.