A novel methodology to classify myoelectric signals using genetic algorithms and support vector machines

The processing of biopotentials is a fundamental part of medical research since it gives the possibility of developing new devices and techniques for the diagnosis, treatment, care, and rehabilitation of patients, in most cases in a non-invasive way. Therefore, creating new methods of analysis for biosignals helps improve the safety, efficiency, and quality of the new processes. Biosignals should be understood as the electrical activity that some cells present, such as cardiac cells, nerve cells, or muscle cells. Due to the above, the study of muscle signals results in one of the most important sources of knowledge about the function of the tissues that make up the system, providing information on the flow of data that crosses the nervous system and the consequent activation of different muscles. This signal can support the detection of some diseases related to electrical activity, such as muscular dystrophy, sclerosis, and neuropathies. Also, this signal can be implemented in control systems, for example, in robotics or prosthetics or even in areas such as telemedicine. EMG signals are very complex, random, non-stationary, non-linear. Therefore, it is necessary to find a pattern that encompasses the signal in general and not the independent data that make it up; due to this, the latest generation EMG pattern recognition systems usually contain blocks of pre-processing, segmentation, extraction of characteristics, dimensionality reduction, classification and data control. The following is a proposal for a methodology based on a support vector machine as a classification algorithm and genetic algorithms as a feature reduction system.