An Architecture of Enhanced EMG Signal Classification based on Multi-Agent System

An Electromyography (EMG) signal provides reliable information and represents an important step towards man-machine communication to address neuro-muscular movements. The vital aspects of this research focus on data acquisition, signal pre-processing, feature extraction, and accurate classification to meet the practical challenges in terms of reliability. The recorded signals first undergo pre-processing in which the uploaded EMG signals are first segmented into clusters using K-means followed by a Genetic Algorithm (GA) that precisely extracts the rows that exhibit potential signal data. The multi-agent system is designed using a Support Vector Machine (SVM) with neural network architecture. This information of trained features deduced using SVM kernel function, also known as Support Vectors (SVs), is fed to Neural Network (NN) for the classification of EMG signals into two classes, namely, normal and pain. The performance of the proposed classification work is evaluated against the Electro- Myography-EMG dataset in terms of precision, sensitivity, f- measure, and accuracy. The simulation analysis over 1000 simulation rounds had demonstrated an average precision of 0.931, sensitivity of 0.915, f-measure of 0.923, and classification accuracy of 95.51%. The results illustrate a better classification accuracy when SVM support vectors are used by FFBPNN to learn the features of normal, and pain class in comparison to the existing EMG signal classification approaches.