Development of Machine-Learning Algorithms for Recognition of Subjects' Upper Limb Movement Intention Using...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalogram (EEG) signals are crucial for understanding brain activity.
• Classifying motor execution and imagination is vital for brain-computer interfaces.
• Eye-blinking artifacts can significantly contaminate EEG data, impacting classification accuracy.

Purpose of the Study:

• To develop and evaluate machine learning algorithms for classifying rest, motor execution (ME), and motor imagery (MI) using EEG signals.
• To assess the effectiveness of Independent Component Analysis (ICA) in artifact reduction for EEG-based movement classification.
• To compare the performance of different machine learning models and window sizes for optimal classification accuracy.

Main Methods:

• Utilized EEG data from fifteen healthy subjects during rest, ME, and MI tasks.
• Preprocessed EEG signals using Independent Component Analysis (ICA) to mitigate eye-blinking artifacts.
• Implemented and compared five machine learning algorithms: KNN, LDA, NB, SVM, and RF, using sliding window techniques (1-2s) and majority voting (MV).

Main Results:

• ICA preprocessing improved motor imagery classification accuracy by up to 6%.
• Majority voting further enhanced accuracy by 1-2% (p<0.05), with greater impact on MI than ME.
• Random Forest (RF) and Support Vector Machine (SVM) demonstrated superior performance for both ME and MI.
• A 2-second window size with RF yielded the highest classification accuracies.

Conclusions:

• ICA and MV strategies are effective in improving EEG signal classification accuracy, particularly for motor imagery.
• RF and SVM are robust algorithms for classifying upper limb movements and intentions from EEG data.
• Optimized window size and preprocessing techniques enhance the reliability of EEG-based brain-computer interfaces.