Assessment of preprocessing on classifiers used in the p300 speller paradigm

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Artifact removal is crucial for accurate electroencephalogram (EEG) signal analysis.
• Raw EEG data requires preprocessing for effective feature extraction and classification.
• The choice of preprocessing methods significantly impacts EEG analysis outcomes.

Purpose of the Study:

• To implement and evaluate three distinct artifact removal techniques for EEG data.
• To assess the influence of different filtering methods on classifier performance.
• To investigate the interdependence between preprocessing strategies and classifier selection in EEG analysis.

Main Methods:

• Implementation of bandpass digital filtering, median filtering, and the facet method for EEG artifact removal.
• Utilizing the BCI competition 2003 dataset for training and testing classification models.
• Comparative analysis of classifier performance across different preprocessing pipelines.

Main Results:

• Achieved classification accuracies ranging from 80% to 96%.
• Demonstrated that optimal classifier performance is dependent on the chosen preprocessing method.
• Two distinct approaches, each with a unique classifier and preprocessing method, reached 96% accuracy.

Conclusions:

• The selection of preprocessing methods and classifiers for EEG analysis are interdependent.
• Tailoring preprocessing techniques to specific classifiers can significantly enhance performance.
• The proposed method, using only three electrodes, offers a cost- and time-efficient approach to EEG measurement.