Optimized Time Window Segmentation for Motor Imagery EEG Classification: An Ant Colony Optimization-Based Approach
Rosi Indah Agustin, Pisut Raphisak · 2025
Motor imagery classification is significantly influenced by the characteristics of individual users. Most state-of-the-art MI-BCI systems rely on fixed time window parameters for EEG data to extract MI features, neglecting individual time latency. In this study, we explore the ant colony optimization (ACO) approach to find the most optimum time window parameters to enhance the feature relevance and improve classification performance. The proposed method comprises a common spatial pattern (CSP) to filter and extract features and linear discriminant analysis (LDA) to classify the subject's intention. Each subject epoch is trimmed based on ACO's best solutions. The proposed method is implemented in BCI Competition IV dataset 2a. Experimental results demonstrate that our proposed approach significantly improves classification accuracy, with mean accuracy gains of up to 10.39% in the mu band and 8.97% in the beta band on the validation set over the fixed-time method. Compared to the random search method, the ACO-based approach achieved a 2.41% improvement in the mu band, with convergence occurring faster in most cases. This study highlights the potential of optimizing time segmentation in EEG-based BCIs using the ACO approach to enhance the classification accuracy.