Expert System Design Based on Wavelet Transform and Non-Linear Feature Selection

Abstract

Electromyogram (EMG) is the record of the electrical excitation of the skeletal muscles, which is initiated and regulated by the central, and peripheral nervous system. EMGs have non-stationary properties. EMG signals of isometric contraction for two different abnormalities namely ALS (Amyotrophic Lateral Sclerosis) which is coming under Neuropathy and Myopathy. Neuropathy relates to the degeneration of neural impulse whereas myopathy relates to the degeneration of muscle fibers. There are two issues in the classification of EMG signals. In EMG’s diseases recognition, the first and the most important step is feature extraction. In this paper, we have selected Symlet of order five of mother wavelet for EMG signal analysis and later six non-linear features have been used to classify using Support Vector Machine. After feature extraction, feature matrix is normalized in order to have features in a same range. Simply, linear SVM classifier was trained by the train–train data and then used for classifying the train-test data. From the experimental results, Lyapunov exponent and Hurst exponent is the best feature with higher accuracy comparing with the other features, whereas features like Capacity Dimension, Correlation Function, Correlation Dimension, Probability Distribution & Correlation Matrix are useful augmenting features.