A Noble EEG Classification Method Using Wavelet Scattering of Feature Extraction for Mental Attention State Detection

The mental attention state directly affects an individual's learning ability. It is important for teachers to understand a student's mental attention state in real teaching. Although many experts have started relevant research on mental attention state recognition, the classification algorithms proposed by these studies can only be limited to a specific field, lacking applicability to common application scenarios. Additionally, the number of mental attention states that can be classified is generally dominated by 2 or 3 categories. Therefore, this paper proposes a general algorithm for mental attention state classification by using electroencephalogram (EEG) signals. First, a wavelet scattering network is introduced to obtain the wavelet scattering features matrix of a specific time segment EEG signal. Feature vectors are obtained by statistical analysis of the wavelet scattering features matrix. The classification model is an ensemble bagged tree; 75% of the feature vectors are used for training the model, while 25% are used as the test set. The experimental results of three open datasets show that the proposed method can better achieve EEG mental attention state classification, with recognition accuracy rates of 99.99%, 100%, and 88.43% in 3, 4, and 5 categories, respectively. Compared with other literature methods, this method has the advantage of high recognition accuracy and reduces the time required for recognition.