การตรวจจับประเภทของอารมณ์จากสัญญาณคลื่นไฟฟ้าสมองด้วยวิธีสตริงแกรมมาร์

Abstract

This research aims to develop a technique for classifying human emotions from electroencephalography (EEG) signals, which are inherently complex and significantly affected by noise. The technique employs string grammar in conjunction with the K-Nearest Neighbors (K-NN) algorithm to enhance the accuracy of emotion classification. Standard datasets, including SEED, SEED-IV, and DEAP, were utilized in this study. These datasets underwent decomposition into low and high frequencies using the Discrete Wavelet Transform (DWT) and statistical moment calculations to generate feature vectors. Subsequently, the feature vectors were clustered using K-means clustering and then converted into character strings for emotion classification with the K-NN algorithm. The experimental results indicate that the developed model achieved a maximum accuracy of 93.75% in Experiment 1 when using the SEED-IV dataset. In Experiment 2, the SEED dataset achieved a maximum accuracy of 71.85%, the SEED-IV dataset achieved a maximum accuracy of 80.56%, and the DEAP dataset achieved a maximum accuracy of 27.73%. In Experiment 3, the SEED dataset reached a maximum accuracy of 71.85%, the SEED-IV dataset reached a maximum accuracy of 79.17%, and the DEAP dataset reached a maximum accuracy of 28.91%. These findings demonstrate the effectiveness of the proposed technique in classifying emotions from EEG signals. However, this method can only work well with data set that is similar to SEED data set