Fuzzy support vector machine with joint optimization of genetic algorithm and fuzzy c-means.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Motor imagery electroencephalogram (MI-EEG) is crucial for neurorehabilitation.
• Fuzzy support vector machines (FSVM) are commonly used classifiers.
• Fuzzy c-means (FCM) is used for membership calculation but is sensitive to initial values and prone to local optima.

Purpose of the Study:

• To enhance the robustness of fuzzy memberships to initial cluster centers in FSVM.
• To improve the classification performance of MI-EEG data.
• To introduce a joint optimization of genetic algorithm (GA) and FCM for an improved FSVM (GF-FSVM).

Main Methods:

• Feature extraction from MI-EEG using improved refined composite multivariate multiscale fuzzy entropy.
• Feature fusion to create a feature vector for each trial.
• GA optimization of FCM initial cluster centers for fuzzy membership calculation and two-class MI-EEG classification.

Main Results:

• High average recognition accuracies of 99.89% and 98.81% on two public datasets.
• Corresponding high kappa values of 0.9978 and 0.9762.
• Demonstrated stability of optimized FCM cluster centers via GA.

Conclusions:

• The GA-optimized FCM cluster centers exhibit significant stability.
• The proposed GF-FSVM achieves superior classification accuracy and consistency for MI-EEG data.
• This method offers a more robust approach for MI-EEG classification in neurorehabilitation.