Abstract—Depression, different from usual mood fluctuations and short-lived emotional responses to challenges in everyday life, is a common illness worldwide, with more than 300 million people affected. Although there are known, effective treatments for depression, fewer than half of those affected in the world (in many countries, fewer than 10%) receive such treatments. The diagnose of depression is usually subject to doctors due to the lack of biomarkers of depression. Electroencephalogram (EEG) is an easy-to-use, cost-effective technique that records electrical activity in brain. In this study, 64-channel EEG data was collected from 213 subjects including 71 health controls and 142 depression patients. 13 different features were extracted from EEG signals from all 7 sub-bands of all channels. 3 different feature selection models were used to find the subset of features that best represents the characteristics of EEG signal and 6 machine learning models were applied on all subsets of features to find the model that gained the highest accuracy and recall on depression detection.