Downsampling and geometric feature methods for EEG classification tasks with CNNs

Anonymous

Keywords: persistent homology, laplacian, eigenvalues, time series, downsampling

TL;DR: We introduce "persistent Laplacian eigenvalues", and evaluate how they perform in time series classification tasks compared to PH and CNNs, across multiple downsampling methods.

Abstract: We experimentally investigate a collection of feature engineering pipelines for use with a CNN for classifying electroencephalogram (EEG) time series from the Bonn dataset. We compare $\epsilon$-series of Betti-numbers and $\epsilon$-series of graph spectra (a novel construction)---two topological invariants of a latent geometry of the timeseries---to raw time series of the EEG to fill in a gap in the literature for benchmarking. Additionally, we test these feature pipelines' robustness to downsampling and data reduction. This paper seeks to establish clearer expectations for both time-series classification via geometric features, and how CNNs for time-series respond to data of degraded resolution.

Previous Submission: No