SCALE-BOSS: A framework for scalable time-series classification using symbolic representations

Time-Series Classification (TSC) is an important problem in many fields across sciences. Many algorithms for TSC use symbolic representation to combat noise. In this paper we propose a framework, namely SCALE-BOSS, to build TSC algorithms that exploit timeseries models based on symbolic representations. While alternative symbolic representations can be incorporated, we have opted to use the Bag-Of-SFA (BOSS) approach, and thus SFA, as a state-of-the-art symbolic time series representation. We investigate the efficiency of several instantiations of this framework based on two main variations, where the TSC model is built either by a time-series classification or by a clustering algorithm. The objective is to advance the computational efficiency of TSC classification algorithms without sacrificing their accuracy. We evaluate the instantiations of the SCALE-BOSS framework on those datasets in the UCR timeseries repository that include the largest training sets. Comparisons with state of the art methods on TSC show the balance between computational efficiency and accuracy on predictions achieved.