Detecting Various Hand Gestures from EMG Signals Using Inkjet-printed Dry Flexible Electrodes with Machine Learning Algorithms

The Electromyography (EMG) signals are one of the most used biological signals which are essential for identifying muscular abnormalities that cause a variety of diseases. Many contemporary technologies are utilized to examine muscle activities through EMG signals. This paper describes a novel dry EMG electrode fabricated with the inkjet-printing (IJP) technique of silver (Ag) nanoparticles on flexible polyamide (PI) film. The design is made up of basic electronic components that are affordable and easy to find. Current research on EMG-based movement classification encounters difficulties with inaccurate classification, lacking generalization ability, and poor robustness. This paper discusses the classification of EMG signals by creating and optimizing various machine-learning models to tackle these issues. For the recognition of different hand gestures based on EMG signals, we focus mainly on 3 areas: First, the preprocessing of the EMG signal obtained from the IJP electrodes, the next area is the extraction of various features in the time domain, finally the recognition of the gestures in the signals using the following classifiers: SVM (Support Vector Machine), Random Forest (RF), Decision Tree (DT) and Artificial Neural Network (ANN). We assessed the ML models’ performance using a dataset containing 38,759 sample values. The RF model in the proposed system reached a top accuracy of 99%. The outcomes of the other classifiers were as follows: Artificial Neural Network (97%), Decision Tree (94%), and Support Vector Machine (90%). This suggested approach is very precise and cost-effective for clinics, making it a possible tool for clinical support in diagnostics and resource optimization.