Building an Adaptive Model of Neural Networks to Extract and Classify EEG Signals
Hoang Thuy Tien Vo, Thi‐Nhu‐Quynh Nguyen, Tuan Van Huynh · 2023
Identifying actions via brainwave signals remains a challenge for researchers because of the limitations in the collection equipment and the data's complexity. Under the advancement of technology, brain waves were more easily collected and stored through sensors connected to computers. One of the types of charts that show brainwave properties is the EEG. Data-01 and Data-02 are used for classification; Data-01 includes emotional states divided into four labels: sad, alert, calm, and happy; Data-02 includes eye behaviors, facial expressions, and wildly imaginative signals divided into 11 labels. The EEG signals are subdivided into five sub-bands, alpha, beta, gamma, theta, and delta, using digital filtering methods such as FIR, IIR, and wavelet transform filters to identify signal characteristics. The neural network method was applied, and Bayesian optimization technology proposed an adaptive neural network. An adaptive neural network can generalize itself as a problem, and change the model, learning rate, and adapt to the input data as needed. The classification model is designed by examining the number of hidden layers, the number of hidden nodes, the learning rate, the transfer functions, and the solver functions. The neural network model's performance is evaluated based on the training performance, and the accuracy of classification results is more than 70%.