Decoding semantic relatedness and prediction from EEG: A classification method comparison.

Area of Science:

• Cognitive Neuroscience
• Machine Learning
• Neuroimaging

Background:

• Machine learning (ML) is crucial for analyzing electroencephalogram (EEG) data in cognitive neuroscience.
• A quantitative comparison of major ML classifiers for EEG decoding in cognition studies is needed.

Purpose of the Study:

• To systematically compare the performance of Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and Random Forest (RF) classifiers.
• To evaluate these classifiers using EEG data from visual word-priming experiments focusing on N400 effects.

Main Methods:

• EEG data from two visual word-priming experiments were analyzed.
• Three ML classifiers (SVM, LDA, RF) were compared using averaged and single-trial EEG data.
• Performance was assessed via decoding accuracy, effect size, and feature importance.

Main Results:

• Support Vector Machine (SVM) demonstrated superior performance compared to LDA and RF.
• SVM outperformed the other methods across all evaluation measures and both experiments.
• The findings highlight SVM's effectiveness in decoding cognitive processes from EEG data.

Conclusions:

• SVM is the most effective ML classifier for decoding EEG data in cognitive neuroscience research, particularly for N400 effects.
• This study provides a quantitative benchmark for selecting ML algorithms in EEG-based cognitive studies.
• The results advocate for the use of SVM in analyzing complex cognitive information from EEG signals.