Surface electromyography based explainable Artificial Intelligence fusion framework for feature selection of hand gesture recognition

Over the past decade, the utilization of machine learning (ML) models for recognizing hand gestures from surface electromyography (sEMG) signals has been in demand for the control of prosthetics. Such real-time control demands swift responses and efficient models. The model's efficiency heavily relies on selecting optimal handcrafted features. As the number of acquisition channels increases, the complexity of handcrafted features escalates the computational burden and necessitates the reduction of irrelevant or noisy features to enhance model efficiency. This study proposes an eXplainable Artificial Intelligence (XAI) fusion-based feature selection framework for sEMG-based hand gesture recognition (HGR). The proposed framework comprises two stages. Firstly, it combines three feature selection methods: filter, wrapper, and embedded. Secondly, it employs SHAP-based explainability for ML models to select relevant features. The first stage of feature selection retains ten relevant features, followed by a fine selection of 50% of features in the second stage. The performance of the proposed framework is compared with three baseline models. It shows that the proposed model performs better than other baseline models in terms of accuracy, precision, recall, f-score, and computational time. The proposed XAI fusion framework achieves an average classification accuracy of 80.75% with the Extra Tree classifier, with a computational time of 3.47 ms. Furthermore, to assess its robustness against baseline models, evaluation is conducted using a publicly available dataset, revealing superior performance compared to other baselines.