Mixture of experts extra tree-based sEMG hand gesture recognition

Abstract Human hand gesture recognition has emerged as a significant research interest in robotic hand control using surface electromyography. It is popular due to its non-invasive nature and ability to capture gesture movements. However, proper handling of overfitting and biases in gesture classification with multiple gestures is a common issue in the development of a generalized classifier. To address this issue, a Mixture of Experts Extra Tree classifier proposes to identify hand gestures. This model utilizes the concept of a Mixture of Experts, which combines individual highly accurate machine learning models (Extra Tree), referred to as experts, each focusing on a specific class, and a fully trained model for all classes, known as the gate (Extra Tree model), is employed to weigh the output of individual expert models. In this study, four subjects performing six hand gesture movements were collected, and their identification was evaluated among eleven models, including the Mixture of Experts Extra Tree classifier. The model was also evaluated on a publicly available dataset containing fifteen gesture classes. Results demonstrate that the Mixture of Experts Extra Tree classifier outperformed the other algorithms considered in this study, accurately identifying hand gesture movements. The MEET classifier achieved accuracies of 86.8%, 89.2%, 87.9%, and 78.4% for four subjects on collected data, and improved mean accuracy by 1.25% on the public dataset.