Ensemble Learning Approach for Subject-Independent P300 Speller

Area of Science:

• Biomedical Engineering
• Neuroscience
• Computer Science

Background:

• Brain-computer interfaces (BCI) enable communication for individuals with severe motor impairments.
• P300 spellers utilize electroencephalography (EEG) signals, specifically the P300 event-related potential (ERP), for character selection.
• Variability in P300 amplitude and latency across individuals necessitates robust BCI training methods.

Purpose of the Study:

• To develop and evaluate ensemble learning classifiers for a P300 speller system.
• To enhance the accuracy and robustness of P300 spellers using generic training (GT) across diverse subjects, including those with amyotrophic lateral sclerosis (ALS).
• To compare the performance of individual classifiers (LDA, SVM, kNN, CNN) against their fused ensemble.

Main Methods:

• Ensemble learning classifiers were developed using Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), k-Nearest Neighbors (kNN), and Convolutional Neural Network (CNN).
• Models were trained on EEG data from healthy subjects for generic training (GT).
• Performance was evaluated by testing on both healthy subjects and ALS patients.

Main Results:

• The fusion of LDA, kNN, and SVM classifiers achieved the highest accuracy.
• An accuracy of 99% was obtained for healthy subjects.
• An accuracy of approximately 85% was achieved for ALS patients, demonstrating effectiveness in a clinical population.

Conclusions:

• Ensemble learning, particularly the combination of LDA, kNN, and SVM, offers a robust approach for P300 speller BCI systems.
• This method improves communication accuracy for individuals with paralyzing disorders like ALS.
• Generic training with ensemble classifiers shows promise for widespread BCI application.