A Significantly Faster Elastic-Ensemble for Time-Series Classification
The Elastic-Ensemble  has one of the longest build times of all constituents of the current state of the art algorithm for time series classification: the Hierarchical Vote Collective of Transformation-based Ensembles (HIVE-COTE) . We investigate two simple and intuitive techniques to reduce the time spent training the Elastic Ensemble to consequently reduce HIVE-COTE train time. Our techniques reduce the effort involved in tuning parameters of each constituent nearest-neighbour classifier of the Elastic Ensemble. Firstly, we decrease the parameter space of each constituent to reduce tuning effort. Secondly, we limit the number of training series in each nearest neighbour classifier to reduce parameter option evaluation times during tuning. Experimentation over 10-folds of the UEA/UCR time-series classification problems show both techniques and give much faster build times and, crucially, the combination of both techniques give even greater speedup, all without significant loss in accuracy.
KeywordsTime series Classification Ensembles Distance measures
- 2.Chen, Y., et al.: The UEA-UCR time series classification archive (2015). http://www.cs.ucr.edu/~eamonn/time_series_data/
- 6.Keogh, E., Pazzani, M.: Derivative dynamic time warping. In: Proceedings 1st SIAM International Conference on Data Mining (SDM) (2001)Google Scholar
- 11.Schäfer, P., Leser, U.: Fast and accurate time series classification with weasel. In: Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, pp. 637–646. ACM (2017)Google Scholar