Classification of EEG Analysis for Sleep Detection Using Wavelet Transform and Neural Network

This paper develops a drowsiness detecting system by resolving the Electroencephalogram (EEG) signals. The acquired EEG signals are subjected to noise and are removed by subtracting the artifacts from the original EEG recording using Biorthogonal Wavelet Filter. Features are extracted using Discrete Wavelet Transform in particular Daubechies’ wavelet with five-level decomposition is utilized to segregate the signal into five sub-bands, namely, delta, theta, alpha, beta and gamma. The statistical moments such as mean, variance, standard deviation and power of the signal are calculated and stored. These moments serve as an input to the next stage, i.e., system classification. Unsupervised learning using K-means clustering is employed as the classification of the signals are not known. Following this, Support Vector Machine and Pattern Recognition Network are employed for supervised classification. This system provides strong decision making during a real-time sleep detection scenario.