<p>Electromyogram (EMG) signals record muscle activity and can be used to replicate the biomechanics of human movement. By utilizing surface EMG signals, accurate classification of individual and collective finger movements can support various applications. Over the past decades, there have been some challenges in the EMG-based hand gesture recognition (HGR) due to the deficient generalization capability, erroneous classification, and weak strength. To resolve these issues and attain high classification performance, this paper presents a novel approach, BRSBELM, that integrates the battle royale optimization (BRO) and the sparse Bayesian extreme learning machine (SBELM) for hand gesture recognition. The proposed model classifies an EMG dataset, which includes the 8 classes of hand gestures. Initially, the EMG signals are pre-processed with the fourth order Butterworth bandpass filter for noise removal. Then, segmentation is performed using the Shannon Energy Envelope (SEE) approach. After segmentation, time-frequency domain features including Wigner-Ville Transform (WVT), Stockwell-Transform (ST), DWT (discrete wavelet transform), and Synchro-Extracting Transform (SET) are extracted. Based on the hand gestures, BRSBELM identifies the EMG signals. The proposed method is simulated in Matlab software using the EMG Myo-readings dataset. To show the effectiveness of the proposed approach, the performance is estimated and compared with the existing methods. The proposed BRSBELM model achieved the highest accuracy of 0.989, outperforming other models including DNN (0.988), CNN (0.985), ANN (0.981), LSTM (0.977), SVM (0.972), and ELM (0.966). Simulation results illustrate that the BRSBELM approach provides the highest classification accuracy for the identification of the hand gestures compared to other approaches.</p>