Classification of distraction descriptor in EEG signal using topological data analysis

Electroencephalogram (EEG) signals are widely used for monitoring brain activity; however their high-dimensionality, sparsity, and non-stationary pose significant challenges for reliable feature extraction and classification. Topological data analysis, particularly persistent homology, has emerged as a powerful approach for capturing intrinsic geometric and topological structures in complex data. In this work, we propose a comprehensive framework that leverages multiple persistent homology-based descriptors, including persistence statistics, landscapes, silhouettes, images, entropy, and Betti curves, for classifying non-clinical EEG signals recorded under varying distraction conditions. EEG data were collected from 45 subjects performing visual tasks in quiet, low-distraction and high-distraction auditory environments. Experimental results demonstrate that topological features consistently improved classification performance, most prominently in the O2 channel, across multiple classifiers, highlighting their complementary role in enhancing non-medical EEG analysis. H 1 -based descriptors outperformed H 0 features, suggesting stronger alignment with the recurrent phase-space structure of delay-embedded oscillatory EEG signals. The preliminary cross-band analysis further indicates that while both alpha and beta frequency bands benefit from topological representations, alpha-band activity tends to exhibit greater sensitivity to distraction. Comparative evaluation against conventional time-domain, frequency-domain, and time-frequency feature extraction methods reveals the competitive performance of the proposed topological approach. These findings highlight the potential of persistent homology as a viable and interpretable complement to conventional EEG-based distraction analysis, with applications in attention monitoring, brain-computer interfaces, and cognitive workload assessment.