The PSO and the Taylor Expansion Based LSSVM Algorithm for EEG Fatigue Multi-Classification

To solve problems of poor convergence speed and large calculation burden of the traditional least squares support vector machine (LS-SVM) algorithm in EEG classification, this paper investigates a particle swarm optimization and a Taylor expansion RBF Kernel Function based LSSVM (PSO-TKF-LSSVM) algorithm. Fisrtly, the PSO algorithm is used to automatically optimize the hyperparameters σ (kernel function width) and γ (the regularization parameter) in the LSSVM algorithm to improve classification accuracy. Secondly, the first three terms of RBF's Taylor expansion are adopted as a kernel function to decrease the computational burden. Finally, the SEED-VIG fatigue data with 17 channels 5 frequency bands and 4 features are taken as classification signals to conduct simulation experiments. Results show that the PSO-TKF-LSSVM algorithm can increase its convergence speed with little impact on the classification accuracy.