EWAT+: Finding Time Series Discords Based on New Discord Measure Functions

Abstract

In this paper, we introduce a new method, called EWAT + , for finding discords in time series databases. The proposed method takes full advantages of WAT, the discord discovery algorithm proposed by Fu et al., with major improvements based on new discord measure functions which help to set up a range of alternative good orderings for the outer loop of the discord discovery algorithm. In addition, we employ a branch-and-bound search mechanism on augmented trie that is carried out in the inner loop of the algorithm. Our experiments show that EWAT + is highly effective in terms of running time.

Parameter-Free Search of Time-Series Discord

“…A multi-resolution extension of [9] has also been proposed (see [17]). To realize early exit of the for-loops, a heuristic based on symbol frequency was recently proposed [18] . Index-based algorithms rely on a set of tuning parameters for efficient encoding of subsequences.…”