Fig 6. Subsequence anomaly detection.

(a) Score of each subsequence in the sample time series in Fig 2(a). The red points indicate the four lowest scores corresponding to four subsequences with the length of ℓ = 110 starting at different positions i in time series. The exclusion-zones are highlighted in gray. (b) Top-4 abnormal subsequences reported and ranked (1^st − 4^th) by the GraphTS.