Multiclass Support Vector Machines for EEG-Signals Classification

In this paper, we proposed the multiclass support vector machine (SVM) with the error-correcting output codes for the multiclass electroencephalogram (EEG) signals classification problem. The probabilistic neural network (PNN) and multilayer perceptron neural network were also tested and benchmarked for their performance on the classification of the EEG signals. Decision making was performed in two stages: feature extraction by computing the wavelet coefficients and the Lyapunov exponents and classification using the classifiers trained on the extracted features. The purpose was to determine an optimum classification scheme for this problem and also to infer clues about the extracted features. Our research demonstrated that the wavelet coefficients and the Lyapunov exponents are the features which well represent the EEG signals and the multiclass SVM and PNN trained on these features achieved high classification accuracies

I. Introduction

The electroencephalogram (EEG) is a complex and aperiodic time series, which is a sum over a very large number of neuronal membrane potentials. Despite rapid advances of neuroimaging techniques, EEG recordings continue to play an important role in both the diagnosis of neurological diseases and understanding the psychophysiological processes. In order to extract relevant information from recordings of brain electrical activity, a variety of computerized-analysis methods have been developed. Most methods are based on the assumption that the EEG is generated by a highly complex linear system, resulting in characteristic signal features like nonstationary or unpredictability [1]. Much research with nonlinear methods revealed that the EEG is generated by a chaotic neural process of low dimension [2]–[4]. According to these reports, the EEG has a finite noninteger correlation dimension and a positive Lyapunov exponent. Furthermore, the distinct states of brain activity had different chaotic dynamics quantified by nonlinear invariant measures such as correlation dimensions and Lyapunov exponents. In recent years, there has been an increasing interest in applying techniques from the domains of nonlinear analysis and chaos theory in studying the behavior of a dynamical system from an experimental time series such as EEG signals [1] [2]–[4]. The purpose of these studies is to determine whether dynamical measures, especially Lyapunov exponents, can serve as clinically useful parameters. Estimation of the Lyapunov exponents is computationally more demanding, but estimates of these parameters are more readily interpreted with respect to the presence of chaos, as positive Lyapunov exponents are the hallmark of chaos [5].