TOTOPO: Classifying univariate and multivariate time series with Topological Data Analysis
This work addresses time series classification for domains where periodicity and shape are discriminative, but it is incremental as it builds on existing topological data analysis approaches.
The authors tackled time series classification by proposing TOTOPO, a method using topological data analysis to extract descriptors from persistence diagrams, which significantly outperformed existing baselines in accuracy and was competitive with state-of-the-art methods, achieving best results on 20% of univariate and 40% of multivariate datasets.
This work is devoted to a comprehensive analysis of topological data analysis fortime series classification. Previous works have significant shortcomings, such aslack of large-scale benchmarking or missing state-of-the-art methods. In this work,we propose TOTOPO for extracting topological descriptors from different types ofpersistence diagrams. The results suggest that TOTOPO significantly outperformsexisting baselines in terms of accuracy. TOTOPO is also competitive with thestate-of-the-art, being the best on 20% of univariate and 40% of multivariate timeseries datasets. This work validates the hypothesis that TDA-based approaches arerobust to small perturbations in data and are useful for cases where periodicity andshape help discriminate between classes.