Page - 159 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

Table 3. Electrode site used for evaluation Number of features Features 1 F4 2 F4, AF4 3 F4, AF4, T7 4 F4, AF4, T7, Pz 5 F4, AF4, T7, Pz, FC6 32 All electrodes The result of multiple regression analysis based on Table 3 is the following Figure 6. Figure 6. Multiple correlation coefficients of EEGs and dips when changing the number of electrodes 4. Conclusion 4.1. Conclusion We have described a method for reducing electrodes in emotional recognition of depression using EEG based on the dataset DEAP. It was confirmed that the multiple correlation coefficient of EEG data and depression increased as the number of electrodes increased. The relationship between EEG data acquired from the electrode site selected by the method in this research and the feeling of depression was clarified. 4.2. Future tasks In this study, we divided the EEG into three frequency bands α, β, γ and then created a feature. This way of dividing is customarily used in the analysis of electroencephalograms, and it is not necessarily clear whether it is effective in emotion recognition of depression. In the future, it is necessary to investigate how to analyze by dividing frequency bands. The feature quantity is created based on the energy of each frequency band, but after separating into each frequency band, it is necessary to newly consider how to create the feature quantity. In this study, wavelet transformation was used to separate EEG into three frequency bands, but we cannot deeply consider time change of EEGs. It is necessary to prepare feature quantities in a form that does not impair time information. Y.Tsurugasaki etal. / IdentificationofEffectiveEEGElectrodes forDepressionSensing 159