Abstract: The global incidence of Alzheimer’s Disease (AD) is on a swift rise. The Electroencephalogram (EEG) signals is an effective tool for the identification of AD and its initial Mild Cognitive Impairment (MCI) stage using machine learning models. Analysis of AD using EEG involves multi-channel analysis. However, the use of multiple channels may impact the classification performance due to data redundancy and complexity. In this work, a hybrid EEG channel selection is proposed using a combination of Reptile Search Algorithm and Snake Optimizer (RSO) for AD and MCI detection based on decomposition methods. Empirical Mode Decomposition (EMD), Low-Complexity Orthogonal Wavelet Filter Banks (LCOWFB), Variational Mode Decomposition, and discrete-wavelet transform decomposition techniques have been employed for subbands-based EEG analysis. We extracted thirty-four features from each subband of EEG signals. Finally, a hybrid RSO optimizer is compared with five individual metaheuristic algorithms for effective channel selection. The effectiveness of this model is assessed by two publicly accessible AD EEG datasets. An accuracy of 99.22% was achieved for binary classification from RSO with EMD using 4 (out of 16) EEG channels. Moreover, the RSO with LCOWFBs obtained 89.68% the average accuracy for three-class classification using 7 (out of 19) channels. The performance reveals that RSO performs better than individual Metaheuristic algorithms with 60% fewer channels and improved accuracy of 4% than existing AD detection techniques.

Key words: Alzheimer’s Disease')">Alzheimer’s Disease, Brain disorder, Electroencephalogram, Reptile Search Algorithm, Snake Optimizer, Optimization