An Improvement to Emotion Detection in EEG Signals Using Deep Artificial Neural Networks

One of the research areas that in recent years several studies have been performed on it is emotion recognition in the EEG signals. In this study, a 4-layered approach has been provided to improve the emotion detection in EEG signals. In this study, we used DEAP data set. We provided a 4-layered approach as follows: 1- Preprocessing 2- Feature Extraction 3-Dimensionality Reduction 4- Emotion detection. To select optimal choices in some stages of these layers, we’ve done some other experiments. The three different experiments have been done. First, finding the right window in the feature extraction. The results shows that Hamming window was the suitable one. Second, selecting the most appropriate number of filter banks in the feature extraction. The results of this experiment showed that 26 numbers was the most appropriate choice. The third experiment was to detect emotions through the proposed method.The results showed 81.58 percent accuracy for arousal, 79.87 percent accuracy for the valence, 80.35 percent accuracy for the dominance dimensions in 2-classes experiment. For 3-classes experiment the results was 68.54 percent accuracy for arousal 66.31 percent accuracy for the valence, 66.92 percent accuracy for the dominance dimensions. The 7.38 percent accuracy improvement in 2-class experiment and 3.38 accuracy improvement in 3-class experiment. This improvement in valence dimension was 7.54 and 5.21, respectively. It seems that using the proposed method can improve emotion detection in EEG signals