Chapter 4 Time series features

The feasts package includes functions for computing FEatures And Statistics from Time Series (hence the name). We have already seen some time series features. For example, the autocorrelations discussed in Section 2.8 can be considered features of a time series — they are numerical summaries computed from the series. Another feature we saw in the last chapter was the Guerrero estimate of the Box-Cox transformation parameter — again, this is a number computed from a time series.

We can compute many different features on many different time series, and use them to explore the properties of the series. In this chapter we will look at some features that have been found useful in time series exploration, and how they can be used to uncover interesting information about your data. We will use Australian quarterly tourism as a running example (previously discussed in Section 2.5).