Chromatic Alpha Complex Generation for EEG Signal Classification

Electroencephalogram (EEG) is high dimensional complex data resembling the electric conduction of neurons. Ana-lyzing the electric activity and complexity of EEG signals became difficult due to its inherent non-stationarity and non-euclidean manifold nature. These intricate characteristics of EEG make the Machine Learning (ML) model laborious in characterizing and classifying the underlying structural connectivity of neurological disorders. Very few studies are conducted on computational geometry and Topological Data Analysis (TDA) for structural analysis and classification of EEG signals. For the first time, this paper proposes a novel technique to classify neurological disorders through the ensembling approach of computational geometry and TDA. The proposed Voronoi Tessellation and chromatic Alpha Complex (VTAC) framework shows the efficacy of using persistent homological data of chromatic alpha complexes in classification. VTAC framework robustness is assessed with publicly available multiple standard datasets with different specifications. Performance metrics of the proposed approach reveal the gain of 10.5% accuracy over baseline models. This shows the potentiality of classification using structural analysis in increasing the model performance as well as understanding the underlying complexity of non-euclidean manifold modalities.