In social communication, recognizing the feelings of others has an important role. Computers can build better relationships between computers and human by recognizing human emotions. Due to depressed feeling, various negative effects such as shortage of sleep, loss of appetite, lowering of concentration ability are concerned. In order to prevent it and further depression, emotion recognition system is necessary. Electroencephalographs (EEGs) are inexpensive among devices that can measure brain activity and are effective in emotion recognition. Since the EEGs acquired from adjacent electrode site of EEG are similar, it is necessary to consider reduction of electrodes, in order to alleviate the complexity of calculation and to avoid overfitting. Previous research has shown valid electrode sites for emotion recognition of depression using an EEG. However, when using an EEG that can flexibly change the measurement sites or an EEG with a small number of electrodes, it is not clear what kind of electrode sites combination should be used to measure EEGs. In this study, we have examined the combination of EEG channels with high contribution to emotion of depression. We have used the dataset DEAP which has EEG data annotated emotion. Explanatory variables were chosen using LARS to identify the effective channels.