Novel accurate classification system developed using order transition pattern feature engineering technique with...

Area of Science:

• Biomedical Signal Processing
• Machine Learning in Healthcare
• Explainable Artificial Intelligence

Background:

• Accurate classification of electroencephalogram (EEG) and electrocardiogram (ECG) signals is crucial for diagnosing neurological and cardiac conditions.
• Existing methods often lack interpretability, hindering clinical adoption.
• Feature engineering plays a vital role in enhancing classification performance for complex biomedical signals.

Purpose of the Study:

• To develop a novel, explainable feature engineering (XFE) framework for high-accuracy EEG and ECG signal classification.
• To enhance the interpretability of machine learning models in biomedical signal analysis.
• To validate the proposed framework on diverse EEG and ECG datasets.

Main Methods:

• Utilized the Order Transition Pattern (OTPat) feature extractor to capture spatial and temporal patterns from signals.
• Employed cumulative weighted iterative neighborhood component analysis (CWINCA) for feature selection.
• Classified features using a t-algorithm k-nearest neighbors (tkNN) classifier.
• Generated interpretable results using Directed Lobish (DLob) and Cardioish symbolic languages for connectome diagrams.

Main Results:

• Achieved over 95% classification accuracy on multiple EEG and ECG datasets.
• Demonstrated 86.07% accuracy on a challenging 8-class EEG artifact dataset.
• The OTPat-based XFE model provided clear, interpretable connectome diagrams for results visualization.

Conclusions:

• The proposed OTPat-based XFE model offers a robust and highly accurate solution for biomedical signal classification.
• The framework's explainability, through symbolic languages, enhances trust and understanding in clinical applications.
• This approach holds significant potential for advancing diagnostic capabilities in neurology and cardiology.