Time series classification techniques

Time  series classifications can be grouped based on type of discriminatory features  techniques are attempting to find \cite{Bagnall_2016}:
  1. Whole  series: two time series compared either as a vector or by a distance measure that uses all the data
  2. Intervals: rather than use whole series, select one or more phase dependent intervals of  the series
  3. Shapelets: based on finding  short, phase independent, patterns (shapelets)  that define class, but that can appear anywhere in series. Class is distinguished by presence or absence of one or more shapelets anywhere in whole series
  4. Dictionary  based: classification based on histograms  constructed from frequency counts  of recurring patterns
  5. Combinations: class of algorithms based on combining two or more of the above approaches into a single classifier
  6. Model  based: Model based algorithms fit a generative model to each series then measure similarity between series using  similarity between models. Commonly  proposed for tasks other than classification or as  part of a larger classification  scheme and are often not as competitive as other approaches (except for long  series of unequal length)

Feature alignment techniques

Illustration  of Dynamic Time Warping alignment process: