Other

Time series features which do not fall into any of the other categories.

Embedding Distance (embedding_dist)

What it does

first represents the time series in a two-dimensional time-delay embedding space (using a time delay equal to the first zero-crossing of the autocorrelation function). It then computes successive distances between points in this 2D embedding space and analyses the probability distribution of these distances. This feature outputs the mean absolute error of an exponential fit to this distribution. It will give low values to time series where the probability distribution of the distance between consecutive time-series values in the 2D embedding space is well approximated by an exponential distribution.

---