Time series / Benchmark

Time-series classification

Time-series classification archives (UCR 2018 univariate, 128 datasets; UEA 2018 multivariate, 30 datasets); grouping used on this site, not an official suite.

Test split: Each dataset has a single fixed train/test split. UCR: 'From the first iteration, the UCR Archive has had a single predefined train/test split' (Dau et al.); the 2018 release reversed the split of 14 datasets to match the UEA archive, and the authors recommend reporting the default split when resampling. UEA: train/test splits provided for all 30 datasets, retaining donor splits where they existed (Bagnall et al.). Test labels are public in both archives.