Page - 158 - in Intelligent Environments 2019 - Workshop Proceedings of the 15th International Conference on Intelligent Environments

In Table 2, red letters are newly selected feature quantities when combining the number of electrodes. The electrode sites obtained by LARS are shown in the figure below. Figure 5. Electrode sites including a feature quantity selected by using LARS The results showed that the electrode site including the feature quantity with a high contribution to the D was particularly frequent in the right frontal lobe. Sharma and colleagues detected depression and suggested that cerebral right hemisphere alone is sufficient for detection of depression [18]. Henriques and colleagues found that depression patients have abnormality in the frontal lobe and temporal lobe compared with healthy subjects [19]. Davidson and colleagues discovered deactivation of left frontal lobe and activation of alpha waves in depressed patients, and in particular activation of the right frontal lobe [20]. Blackhart and colleagues investigated the fact that people showing activation of electroencephalograms in the right frontal lobe deteriorated symptoms such as depression and anxiety after one year [21]. The result that the electrode site of the right frontal lobe of this study is important is considered to be similar results as compared with the above studies. 3. Evaluation In this study, we investigated combinations of electrode sites with high contribution to emotional recognition of depression. To evaluate the obtained results, we investigated the accuracy of emotional recognition of depression when changing the number of electrodes by multiple regression analysis. At that time, 10-fold cross validation was performed. Fifteen feature values of average, median, variance, maximum value, minimum value in the γ, β and α wave frequency bands were used as the feature quantities used for multiple regression analysis. In addition, when performing multiple regression analysis, the change in the number of electrodes was made based on the electrode obtained as a result of LARS. The correspondence table between the change in the number of electrodes and the electrode site is shown below. Y.Tsurugasaki etal. / IdentificationofEffectiveEEGElectrodes forDepressionSensing158