Motifs for prediction

Question

I'm trying to find a method for using motifs for prediction. Motifs defined as similar (sub)sequences can be found using approximate or exact methods, see Exact Discovery of Time Series Motifs.

My problem is: once you have motifs of labelled time series how could the motifs be used for prediction?

Answer 1

Since a motif is a pattern that gets repeated in a time series if you observe the motif repeating itself after a certain time lag in the series then you can forecast a future occurrence of the motify in the future one time lag ahead of where it last occurred.

Since you clarified that you have labels for your series and based on an observed motif you want to identify the right label, I would say that if each time series has different motifs then observing a particular motif would match it to a particular series and thereby identify the correct label.