Classification of EEG signals from musicians and non-musicians by neural networks

Long-term training will change the brain activity due to plasticity of the human brain. In this paper, an EEG-based neural network was proposed to assess neuroplasticity induced by musical training. A musical interval perception experiment was designed to acquire and compare the behavioral and neural responses of musicians and non-musicians. The auditory event related potentials (AEP) elicited by the consonant and dissonant intervals were combined and the PCA was used to extract discriminable features to classify the EEG recordings. Various linear and nonlinear classifiers were utilized for EEG classification and the results were also compared. The average accuracies of LDA, RBFSVM and BPNN are 94.6% (PCs = 8), 95.9% (PCs = 6), and 97.2% (PCs = 20). ANOVA analysis of the classification results shows that the performance of BPNN is significantly better than the results of LDA (p<;0.05). But there is no significantly difference with RBFSVM. The RBFSVM performs better stability if redundant principle components were included in the feature vector. The experimental results demonstrate the feasibility of assessing effects of musical training by AEP signals elicited by musical chord perception.