2020/10/10 by Polina Pilyugina, Pilyugina, Polina, Rodrigo Rivera-Castro +3
Computer Science · Medicine · #Algorithm #Artificial intelligence #Benchmarking #Cartography #Clusterin in disease pathology #Computer science #Data mining #FOS: Computer and information sciences #Geography #Machine Learning (cs.LG) #Machine learning #Missing data #Multivariate analysis #Multivariate statistics #Pattern recognition (psychology) #Scale (ratio) #Series (stratigraphy) #Time series #Topological and Geometric Data Analysis #Topological data analysis #Univariate #cs.LG
paper · pdf · doi:10.48550/arxiv.2010.05056
published in arXiv (Cornell University) (Cornell University)
arxiv created 2020/10/10 · openalex publication_date 2020/10/10 · arxiv updated 2020/10/13 · openalex created_date 2022/07/25 · openalex updated_date 2026/08/08
This work is devoted to a comprehensive analysis of topological data analysis\nfortime series classification. Previous works have significant shortcomings,\nsuch aslack of large-scale benchmarking or missing state-of-the-art methods. In\nthis work,we propose TOTOPO for extracting topological descriptors from\ndifferent types ofpersistence diagrams. The results suggest that TOTOPO\nsignificantly outperformsexisting baselines in terms of accuracy. TOTOPO is\nalso competitive with thestate-of-the-art, being the best on 20% of univariate\nand 40% of multivariate timeseries datasets. This work validates the hypothesis\nthat TDA-based approaches arerobust to small perturbations in data and are\nuseful for cases where periodicity andshape help discriminate between classes.\n