The development of complex human-machine systems, for example, training complexes, involves the collection and processing of large amounts of data to form biofeedback. One of the data sources is electromyography, which provides information about human muscle activity. The analysis of muscle activity is used in professional, medical virtual simulators, as well as rehabilitation systems. However, in its original form, electromyography data are poorly applicable in real-world tasks, and their processing and analysis are required. This study examines the application of machine learning to classify human muscle activity in various types of human movements. The conducted experimental studies have shown the applicability (up to 100% accuracy) of various machine learning algorithms and neural networks to solve the problem. Several experiments have been conducted on various parts of the human body: leg movement (squatting, bending/unbending the knee, pulling the leg to the side), arm movement (clenching/ unclenching the fist, lifting various loads).