4.7 Further reading

• R. B. Cleveland et al. (1990) introduced STL, and it still provides the best description of the algorithm. The idea of using STL for features originated in Wang et al. (2006).
• The features provided by the feasts package were motivated by their use in Hyndman et al. (2015) and Kang et al. (2017).
• The exporation of a set of time series using principal components on a large collection of features was proposed by Kang et al. (2017).