Unified Multi-Class Electroencephalogram Artifact Recognition Using Machine Learning Classifiers

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Electroencephalographic (EEG) recordings are often contaminated by artifacts, which are noisy signals that can degrade data quality and affect neurophysiological analysis.
• Existing methods often focus on single artifact types, potentially leading to increased false positives in anomaly detection for conditions like epilepsy.
• A unified approach is needed to simultaneously classify various artifacts and brain activity for improved EEG data interpretation.

Purpose of the Study:

• To develop and evaluate a unified multi-class framework for classifying diverse EEG artifact types and brain activity (normal and pathological).
• To assess the performance of simple Machine Learning classifiers within this framework.
• To enhance the reliability and interpretability of EEG data analysis.

Main Methods:

• Extraction of spectral, temporal, and statistical features relevant to different artifact types.
• Training and evaluation of K-Nearest Neighbors (KNN) and XGBoost classifiers using a cross-validation scheme.
• Implementation of a unified framework capable of multi-class classification.

Main Results:

• Both KNN and XGBoost models demonstrated strong performance in classifying multiple EEG artifact types and brain activity.
• Achieved approximately 90% sensitivity across all classification classes.
• Maintained 100% specificity, indicating high reliability in distinguishing between classes.

Conclusions:

• A unified, multi-class framework is effective for classifying various EEG artifacts and brain activity.
• Simple Machine Learning classifiers can achieve high performance in this complex classification task.
• The proposed approach offers an interpretable and robust method for improving EEG data quality and analysis.