Mother Wavelet for Optimal Feature Analysis in Multiclass EEG Signals

Abstract

The aim of this study is to investigate the best type of mother wavelet capable of classifying multiple classes related to EEG. For instance, classification of the three brain states, namely seizure, pre-seizure (for seizure prediction), and normal states is an important part of the study in multiclass classification of epilepsy. In an attempt to yield the best mother wavelet, the study employs the MDWP approach by excavating through the wavelet packet tree up to the seventh level of decomposition, exploiting the wavelet coefficients on each level. The mother wavelets incorporated in the study are the commonly used wavelets, namely db4, sym5, coif4 and db2. Features were obtained by evaluating energy on all wavelet packets, which were further ranked using Naïve-Bayes classifier. Beginning with the feature ranked highest and progressively adding features with lower ranks one at a time, the classification results depicted in the form of patterns show the db4 mother wavelet to outperform others.