A novel AVOA optimized DNN-BiLSTM-attention model for improved gesture classification using electromyography signal
Gautam Shah, Abhinav Sharma, Deepak Joshi, Ajit Singh Rathor · Results in Engineering · 2025
Identification of hand gestures using wearable interfaces has gained significant attention in areas such as human–computer interaction, gaming, sign language recognition, rehabilitation and assistive robotics. This study presents a hybrid deep learning architecture that integrates convolutional layers with BiLSTM and attention mechanisms to classify upper limb movements in healthy individuals using four distinct feature sets (F1, F2, F3, F4). A key contribution of this work is the optimization of convolutional hyperparameters, such as the number of filters, using various nature-inspired metaheuristic algorithms, thereby eliminating the need for manual tuning. Extensive experiments conducted on EMAHA-DB1 dataset demonstrate the effectiveness of the proposed method. Comparative evaluations against several state-of-the-art ML models, namely Cubic SVM, LDA, Decision Tree, and Deep Neural Network, reveal that our model chieves superior performance across all four distinct feature sets. Specifically, the AVOA-optimized DNN-BiLSTM-Attention model achieved classification accuracies of 99.35% for F1, 99.72% for F2, 94.39% for F3, and 99.19% for F4 feature set. Furthermore, statistical analysis confirms the model’s robustness and significant improvements across all feature sets, highlighting its superiority.