Classification of wavelet-denoised musical tone stimulated EEG signals using artificial neural networks

Electroencephalogram (EEG) signals contains information which may be of interest for a certain purpose. However, this information may be clouded by noise. The necessity of extracting this information using filtering and feature extraction techniques is of great importance. In this study, the wavelet de-noising was implemented instead of the usual frequency filter methods. Daubechies (usually denoted by `db') wavelets (`db1' to `db10') were utilized to determine if wavelet-based de-noising is effective in preparing musical tone stimulated EEG signals for feature extraction leading to classification. The selection of wavelet is based on signal-to-noise ratio (SNR), peak signal-to-noise ratio (PSNR), mean square error (MSE) and correlation coefficient (R). Twelve features were used and fed into an artificial neural network for classification. Results show that among the ten wavelets used, `db8', `db9' and `db10' were found to be useful having satisfied the selection criteria. The EEG signals were divided into 5 segments: Baseline, secondary baseline, C, F and G. It was found out that each segment can be classified using different wavelets with correct classification accuracy ranging from 80% to around 92%.