Using machine learning algorithms for grasp strength recognition in rehabilitation planning

Abstract

The augmentation of individuals' quality of life, particularly those with disabilities, can be achieved through state-of-the-art artificial intelligence solutions. Machine learning algorithms, known for their ability to acquiring knowledge and identify significant characteristics from diverse datasets, play a crucial role. In this investigation, we focused on classifying various weights commonly encountered in daily activities based on electromyography (EMG) readings, using multiple distinct machine learning algorithms. This endeavor involved collection of substantial data from a substantial cohort, wherein participants assumed distinct arm configurations while manipulating three various objects (specifically, a pen, a bottle, and a weighty object) or no object at all. The sample encompassed 50 physically capable and healthy participants, with an equal distribution of 25 males and 25 females. The muscular activity was measured utilizing the MYO armband, an advanced eight-channel EMG device positioned on the forearm. After the preprocessing of this data, several machine learning algorithms has been employed to analyze the dataset. Notably, the outcomes demonstrate that the K-Nearest Neighbors (KNN), Random Forest (RF), and Decision Tree (DT) algorithms emerge as the optimal methodologies for grip strength estimation, achieving impressive accuracy rates of 99.23 %, 99.08 %, and 98.62 %, respectively. The experimental data, and supplementary materials are available at https://github.com/arshiaeskandari/EMG-Dataset.