Chaotic Aspects of EMG Signals in Normal and Aggressive Human Upper Arm Actions

Abstract

The aim of this research work is to demonstrate a standardized procedure for extracting subtle chaotic features to study the dynamics of aggressive and non-aggressive human muscle actions in the chaotic domain. The relevant features present in the electromyogram (EMG) signals are analyzed by exploiting the chaotic characteristics of the signal. Degree of Self-Similarity (DoSS), Largest Lyapunov Exponent (LLE), Correlation Dimension (CD), Approximate Entropy (ApEn) and Katz Fraction Dimension (KFD) are the features, extracted to study the chaotic aspects of normal and aggressive human upper arm muscles. This chaotic feature vector is utilized for signal characterization, which is fruitfully extended for classification of the EMG signals into aggressive and normal. The proposed extraction and classification technique was experimentally verified for validating the findings, using EMG signals available from the UCI machine learning repository database. The features are statistically categorized into three significant levels, applying ANOVA technique. The inferences lead us to conclude that the extracted chaotic features qualify as a distinguishing multi-feature set for EMG signals of different classes. Five different classifications approached were used for classification by using tenfold cross-validation. The maximum classification accuracy achieved was 97.5% with two of the most significant chaotic features.