ELIS++: a shapelet learning approach for accurate and efficient time series classification

Abstract In recent years, time series classification with shapelets, due to the high accuracy and good interpretability, has attracted considerable interests. These approaches extract or learn shapelets from training series. Although they can achieve higher than other approaches, there still confront some challenges. First, may suffer low in case of small dataset. Second, must manually set parameters, like number length each shapelet beforehand, hyper-parameters, learning rate regulation weight, which are difficult without prior knowledge. Third, extracting incurs a huge computation cost, search space. this paper, we extend our previous approach ELIS ELIS++. To improve on dataset, propose data augmentation approach. quality based PAA candidates technique proposed ELIS, ELIS++ first novel entropy-based candidate selection mechanism discover candidates, then applies logistic regression model adjust shapelets.To avoid setting parameters manually, Bayesian Optimization Moreover, two techniques efficiency, coarse-grained adjustment SIMD-based parallel computation. We conduct extensive experiments 35 UCR datasets, results verify effectiveness efficiency