Advanced EEG Signal Classification for Neural Prosthetic Devices using Metaheuristic and Deep Learning Techniques

Thippagudisa Kishore Babu¹Damodar Reddy Edla¹Suresh Dara²ALLAM Mohan^2*

• ¹Indian Institute of Technology Goa, Ponda, India
• ²VIT-AP University Campus, Amaravati, India

The classification of high-dimensional biomedical signals, particularly electroencephalography (EEG), continues to face challenges due to redundant and irrelevant features that hinder the performance of learning models. This study proposes a unified framework integrating a Coati optimization algorithm (COA) with machine and deep learning classifiers to enhance EEG-based neural prosthetic control. The novelty of the proposed approach lies in its dynamic, parameter-free feature selection strategy, which introduces diverse candidate solutions across iterations and provides a more adaptive exploration–exploitation balance compared to existing evolutionary and swarm-based methods. The optimized feature subsets are then used to train classifiers including SVMs, Random Forests, CNNs, and RNNs for decoding motor imagery signals. Experimental evaluations on standard EEG datasets demonstrate that the proposed COA-based framework significantly improves performance, achieving a 6.5% increase in classification accuracy over the best competing feature selection method. These results highlight the effectiveness of dynamic evolutionary feature selection in advancing real-time neural prosthetic systems.