Enhancing EEG-Based Mental Stress State Recognition Using an Improved Hybrid Feature Selection Algorithm.

Area of Science:

• Neuroscience and Signal Processing
• Biomedical Engineering
• Machine Learning for Health

Background:

• Wearable electroencephalogram (EEG) devices are crucial for real-life mental stress recognition.
• Efficient EEG channel usage and optimal feature selection are vital for accurate stress detection.
• Existing methods often struggle with high-dimensional feature spaces and suboptimal performance.

Purpose of the Study:

• To identify an optimal feature subset for discriminating mental stress states.
• To enhance the overall classification performance of mental stress recognition systems.
• To develop an efficient feature selection method for wearable EEG applications.

Main Methods:

• Extraction of multi-domain features: time, frequency, time-frequency, and network connectivity.
• Proposed a hybrid feature selection (FS) method: minimum redundancy maximum relevance with particle swarm optimization and support vector machines (mRMR-PSO-SVM).
• Validated the proposed method on four diverse datasets: EDMSS, DEAP, SEED, and EDPMSC.

Main Results:

• The mRMR-PSO-SVM method significantly reduced the feature vector space by an average of 70% compared to state-of-the-art methods.
• Achieved significant increases in overall detection performance for mental stress.
• Demonstrated superior performance and efficiency over existing metaheuristic methods.

Conclusions:

• The proposed hybrid feature selection method is effective for mental stress recognition using EEG signals.
• This approach offers a computationally efficient and highly accurate solution for wearable stress detection systems.
• The findings contribute to the advancement of non-invasive mental health monitoring technologies.