Multi optimized SVM classifiers for motor imagery left and right hand movement identification

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Science

Background:

• Electroencephalography (EEG) signals offer a potential control method for individuals with motor impairments.
• Brain-computer interfaces (BCIs) leverage EEG signals for human-computer interaction, but accurate brain task identification remains challenging.
• Existing BCI methods using EEG signals for motor imagery classification show limited accuracy.

Purpose of the Study:

• To improve the accuracy of identifying motor imagery tasks (left vs. right hand actions) using EEG signals.
• To explore the effectiveness of novel features and optimized Support Vector Machine (SVM) classifiers for BCI applications.
• To enhance classification performance beyond current state-of-the-art methods.

Main Methods:

• Extraction of novel features from EEG signals corresponding to left and right hand motor imagery.
• Feature selection using genetic algorithm optimization.
• Classification using single and multiple optimized SVM classifiers.

Main Results:

• A single optimized SVM classifier achieved a mean classification accuracy of 89.8%.
• A system employing three optimized SVM classifiers with diversified features reached a mean performance of 94.11%.
• This multi-classifier approach significantly outperforms existing methods, which typically do not exceed 81% accuracy on the same dataset.

Conclusions:

• Combining multiple optimized SVM classifiers with carefully selected features is a highly effective strategy for improving BCI performance.
• This optimized multi-classifier approach offers a promising advancement for BCI applications, particularly for individuals with disabilities.
• The proposed method demonstrates a significant leap in EEG-based motor imagery classification accuracy.