Random Walk Binary Grey Wolf Optimization for feature selection in sEMG based hand gesture recognition

In the last few years, there has been a significant on the relation between Human Machine Interface (HMI) and hand gesture recognition. Moreover, researchers are recently focusing further on the development of new methods for gesture classification and signal feature selection mainly based on proposed swarm intelligence optimization methods. This paper presents a versatile optimizer - Random Walk Binary Grey Wolf Optimization. Belonging to the category of evolutionary optimization, this meta-heuristic algorithm can solve optimization problems and perform feature selection. Firstly, the optimizer improvises searchability through random walks in the search space. Secondly, through the optimal selection of features, the optimizer improvises the prediction accuracy and reduces the computational complexity. Moreover, this paper presents a collection of surface electromyography sEMG signals from 10 subjects performing each 10 gestures for 10 observations each. For the data collection, two myoware sensors were used on the forearm of the participants. Features in time and feature domain were extracted from the collected signals and used for the RWBGWO implementation. The performance of the proposed algorithm is verified using the collected sEMG for gesture recognition. The results and observations show that RWBGWO is an efficient optimizer having a better performance in comparison with BGWO, binary particle swarm optimizer BPSO and binary hybrid grey wolf particle swarm optimizer BGWOPSO in feature selection with the least number of selected features achieving a higher classification accuracy reaching 95.5%.