TriPat-XFE: a triangle pattern-based explainable feature engineering framework for EEG classification

Abstract

Background: Electroencephalography (EEG) signals are cost-effective tools for capturing brain activity and are often called the brain's language. Their non-invasive nature and rich information content make them valuable for brain research. Extracting meaningful features from EEG signals is critical for effective analysis and interpretation. New Methods: We introduce a novel feature extraction method called Triangle Pattern (TriPat), designed to provide both accurate and explainable results. Built around this method, we propose a new explainable feature engineering (XFE) framework. This framework uses TriPat for feature extraction, CWINCA for feature selection, and tkNN for classification. The selected features also drive the Directed Lobish (DLob) explainable AI system, which generates symbolic explanations. This unified approach enables both high-performance classification and interpretable outputs. Results: The proposed TriPat-centric XFE framework achieved over 90% classification accuracies across three EEG datasets, including Artifact, Stress, and Psychosis. In addition to high accuracy, the framework produced cortical and hemispheric connectome diagrams that visualize the underlying brain activity patterns. Comparison with existing methods and conclusion: Compared to existing models, TriPat-centric XFE offers higher accuracy with lower computational cost. It operates efficiently on standard hardware without GPUs and produces explainable results using DLob. These advantages make the framework both practical and interpretable for EEGbased brain analysis.