Subject based feature selection for hybrid brain computer interface using genetic algorithm and support vector machine

• Published in: Results in engineering Vol. 27; p. 105649
• Main Authors: Mateen, Nida, Naeem, Mehreen, Khan, Muhammad Jawad, Yousaf, Talha, Ali, Ahsan, Altabey, Wael A., Noori, Mohammad, Kouritem, Sallam A
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2025; Elsevier
• Subjects: Brain–computer interface; EEG-EMG; EEG-fNIRS; Genetic algorithm; Hybrid modalities; Subject-specific feature selection; Support vector machine; EEG-EMG; Genetic algorithm; EEG-fNIRS; Subject-specific feature selection; Brain–computer interface; Support vector machine; Hybrid modalities
• ISSN: 2590-1230; 2590-1230
• DOI: 10.1016/j.rineng.2025.105649

•Devised a modified version of genetic algorithm for feature selection of hybrid BCIs.•Used SVM as an objective function for GA.•To verify the effectiveness of proposed model, used two online available data.•Accuracy across hybrid BCI datasets was greatly enhanced by using the proposed genetic algorithm.•The approach’s effectiveness is shown by the 4 % and 5 % average accuracy gains for EEG-EMG and EEG-fNIRS, respectively. Feature selection is of great importance in hybrid BCI systems to reduce dimensionality, improve interpretability, and optimize the classification performance. In this research, we present a subject-specific feature selection technique based on the modified genetic algorithm (GA) and support vector machine (SVM) classifier. The GA includes an explored list feature and logical checkpoints to prevent premature convergence and efficiently search the space of features. Experiments were performed on publicly available hybrid EEG-EMG and EEG-fNIRS datasets. Evaluation of the proposed method was performed for different channel counts, window frame sizes and lengths of feature combination (2, 3 and 4). On average, classification accuracy improved by 4 % and 5 % for EEG-EMG and EEG-fNIRS modalities, respectively, compared to baseline. The framework outperforms traditional filter- and wrapper-based feature selection methods on representative subjects, confirming its robustness and adaptability across individual neural patterns. These results highlight the importance of personalized feature selection in hybrid BCIs and demonstrate the viability of evolutionary algorithms for real-time, low-latency brain–machine applications.