Gesture recognition and classification based on Newton Raphson optimised XGBoost algorithm

Abstract

In this study, the XGBoost algorithm is improved using the state-of-the-art Newton-Raphson Optimisation Algorithm (NRBO) for the detection of EMG signals and for gesture recognition and classification. On the training set, we observed four gesture types: stationary hand, clenched fist hand, wrist flexion, and wrist extension, all of which achieved 100% accuracy in prediction. While a total of 210 gestures were correctly predicted on the test set, only three gestures were incorrectly predicted. Specifically, a hand that was supposed to be stationary was mispredicted as clenched fist, a gesture that was supposed to be wrist flexion was mispredicted as stationary, and a hand that was supposed to be stationary was mispredicted as wrist extension. Overall, our proposed XGBoost model based on NRBO optimisation exhibits 98.59% accuracy and performs well in gesture prediction and classification. This study is significant, not only improving the accuracy and efficiency of EMG signal processing techniques, but also providing useful insights for future research in related fields.