Optimizing deep CNN architecture via hybrid Harris Hawks arithmetic algorithm for EEG meditation classification

Meditation is a widely recognized practice that enhances mental well-being and cognitive function. Despite advances in EEG meditation neuroscience, challenges persist in extracting robust and interpretable features from complex, non-stationary EEG signals. Existing classification methods often rely on limited feature sets and traditional machine learning approaches. These methods lack comprehensive integration of advanced time-frequency analysis, deep learning, and modern nature-inspired optimization techniques. To address this gap, we introduce a hybrid EEG-based theta-band meditation classification framework that combines Harris Hawks Optimization (HHO) and the Arithmetic Optimization Algorithm (AOA) to tune the parameters of a Convolutional Neural Network (CNN). EEG signals are pre-processed and converted into time-frequency images using the Stockwell Transform (S-transform). These images are fed into the proposed HHO-AOA-CNN framework, where HHO explores and AOA exploits to achieve effective hyper-parameter optimization. The optimized CNN is then used to classify EEG data into three categories: Vipassana (VIP), Isha Shoonya (IS), and Control (CTR). Experimental results demonstrate that the hybrid model outperforms standalone HHO-CNN, AOA-CNN, and baseline CNN models. The proposed approach achieves an accuracy of 94.20%, indicating strong classification performance. Additionally, statistical measures such as best, worst, average fitness, and standard deviation confirm the stability and robustness of the hybrid optimizer.