Comparative Analysis of EEG Sub-band Powers for Emotion Recognition

For several years, emotion recognition systems have been an attraction for the manual feature engineering field, which is detected as the most prominent feature in classification problems. Frequency-based analysis has been successfully used in emotion recognition systems. However, all relevant and irrelevant frequency bands are evaluated using the testing procedure. The aim of this study is to perform a comprehensive analysis of sub-band frequency regions between each other and to elicit the most crucial frequency band. Highly preferred SEED and DEAP datasets are used to achieve this purpose. Three different traditional machine learning methods, such as kNN, RF, and SVM, are used as a classifier. Experimental studies have been carried out, and for the SEED dataset, the gamma band provided 92.56% success rate, for the DEAP dataset, 80.81% and 81.57% success rates were presented for the valence and arousal axes, respectively. The benefit of this study is that it is more helpful to operate with only a few frequency regions instead of operating with all sub-frequency regions. In this way, only band-pass filters may be sufficient instead of the signal processing techniques such as FFT and Wavelet, which is the first stage of EEG-based emotion recognition.