Motif Detection Inspired by Immune Memory

Abstract

The search for patterns or motifs in data represents an area of key interest to many researchers. In this paper we present the Motif Tracking Algorithm, a novel immune inspired pattern identification tool that is able to identify variable length unknown motifs which repeat within time series data. The algorithm searches from a neutral perspective that is independent of the data being analysed and the underlying motifs. In this paper we test the flexibility of the motif tracking algorithm by applying it to the search for patterns in two industrial data sets. The algorithm is able to identify a population of meaningful motifs in both cases, and the value of these motifs is discussed.

Motif detection inspired by immune memory

“…The MTA is used to find motifs in this power demand data and to identify inconsistencies in the periodicity of those motifs [21]. The inconsistencies could be indicative of potential anomalies in the data such as bank holidays or periods where the research centre had to close unexpectedly.…”

Motif Detection Inspired by Immune Memory

“…From an initial scan by eye it is unclear whether any significant motifs exist, representing an ideal challenge for the two algorithms. The parameters for the MTA for this data set have been selected based on the results of previous testing [22,21]. An alphabet size a=6 was set in accordance with that selected by Keogh.…”

“…A bind threshold of r = 0.15 is selected to ensure only close fitting motifs are identified. For an analysis of the sensitivity of the MTA to changes in these parameters readers are directed to the analysis in our earlier work [22,21].…”

“…In the second scenario the MTA is tested on the power demand data set [21]. This is a publicly available data set 1 containing 35,040 data values.…”

“…An alphabet size a = 6 is selected based on results of previous testing [22,21]. The symbol size is increased to s = 250 due to the increased size of the power demand data set.…”