eduzhai > Applied Sciences > Engineering >

Interpretable Time Series Classification using Linear Models and Multi-resolution Multi-domain Symbolic Representations

  • Save

... pages left unread,continue reading

Document pages: 31 pages

Abstract: The time series classification literature has expanded rapidly over the lastdecade, with many new classification approaches published each year. Priorresearch has mostly focused on improving the accuracy and efficiency ofclassifiers, with interpretability being somewhat neglected. This aspect ofclassifiers has become critical for many application domains and theintroduction of the EU GDPR legislation in 2018 is likely to further emphasizethe importance of interpretable learning algorithms. Currently,state-of-the-art classification accuracy is achieved with very complex modelsbased on large ensembles (COTE) or deep neural networks (FCN). These approachesare not efficient with regard to either time or space, are difficult tointerpret and cannot be applied to variable-length time series, requiringpre-processing of the original series to a set fixed-length. In this paper wepropose new time series classification algorithms to address these gaps. Ourapproach is based on symbolic representations of time series, efficientsequence mining algorithms and linear classification models. Our linear modelsare as accurate as deep learning models but are more efficient regardingrunning time and memory, can work with variable-length time series and can beinterpreted by highlighting the discriminative symbolic features on theoriginal time series. We show that our multi-resolution multi-domain linearclassifier (mtSS-SEQL+LR) achieves a similar accuracy to the state-of-the-artCOTE ensemble, and to recent deep learning methods (FCN, ResNet), but uses afraction of the time and memory required by either COTE or deep models. Tofurther analyse the interpretability of our classifier, we present a case studyon a human motion dataset collected by the authors. We release all the results,source code and data to encourage reproducibility.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...