Least square support vector machine based Multiclass classification of EEG signals
Deepesh Kumar, Rajesh Kumar Tripathy, Ashutosh Acharya · 2014
This paper describes the pattern recognition technique based on multiscale discrete wavelet transform(MDWT) and least square support vector machine (LS-SVM) for the classification of EEG signals. The different statistical features are extracted from each EEG signal corresponding to various seizer and nonsiezer brain functions, using MDWT. Further these sets of features are fed to the LS-SVM multiclass classifier for the classification. At the output, the required classifier predicts the output level corresponding to the given test features. The actual output levels are compared with the classifier's predicted output levels and the performance of classifier determined using classification rate (CR). The outcome of our result confirms that the LS-SVM multiclass classifier with linear kernel function, One VS All coding algorithm and 10 fold cross validation scheme gives better performance in terms of CR of 98.07% than other algorithm based LS-SVM multiclass classifier for the required EEG signal classification.