Related Experiment Video
Updated: Feb 26, 2026
07:37
Assessment and Communication for People with Disorders of Consciousness
Published on: August 1, 2017
9.7K
A review and experimental study on the application of classifiers and evolutionary algorithms in EEG-based
Farajollah Tahernezhad-Javazm1, Vahid Azimirad1, Maryam Shoaran1
1Department of Mechatronics, The Center of Excellence for Mechatronics, School of Engineering Emerging Technologies, University of Tabriz, Tabriz, Iran.
Journal of Neural Engineering
|July 19, 2017
Summary
This study surveys classification and evolutionary methods for electroencephalography (EEG) brain-machine interface (BMI) systems. Linear discriminant analysis, support vector machines, and invasive weed optimization algorithms showed the best performance for EEG signal classification.
Area of Science:
- Biomedical Engineering
- Machine Learning
- Neuroscience
Background:
- Noninvasive brain-machine interface (BMI) systems are crucial for future human-computer interaction.
- Electroencephalography (EEG) signals are widely used in BMI systems due to their noninvasiveness.
- Developing effective classification and optimization methods for EEG signals is essential for advancing BMI technology.
Purpose of the Study:
- To provide a comprehensive theoretical and experimental survey of classification and evolutionary methods for EEG-based BMI systems.
- To review and investigate various base and combinatorial classifiers, including boosting and bagging techniques, and evolutionary algorithms.
- To assess and compare the performance of these methods on established BMI paradigms.
Main Methods:
- Reviewed and investigated base classifiers (e.g., linear discriminant analysis, support vector machines, naive Bayes) and combinatorial classifiers (e.g., bagging decision trees, logistic regression, adaptive boosting).
- Assessed and compared classifiers and evolutionary algorithms using sensory motor rhythm-BMI and event-related potentials-BMI systems.
- Experimentally evaluated improved evolutionary algorithms and bi-objective algorithms using cross-validation accuracy (CVA) and stability to data volume (SDV) as criteria.
Main Results:
- Linear discriminant analysis and support vector machines achieved the best CVA for base classifiers.
- Naive Bayes demonstrated the best stability to data volume (SDV) among base classifiers.
- Invasive weed optimization (IWO) and its bi-objective variant showed superior performance among evolutionary algorithms.
Conclusions:
- The study provides a valuable overview of classification and optimization techniques for EEG-based BMI.
- Experimental results highlight the effectiveness of specific classifiers and evolutionary algorithms for different performance metrics.
- The findings contribute to the advancement of noninvasive BMI system development and application.
