Mining undemanding and intricate patterns with periodicity in time series databases

Abstract

Existing periodic pattern mining algorithms detect only specific type of periodicity, i.e., either symbol, sequence, or segment periodicity. Thus, the user is not able to find the combination of any two periodicities using existing algorithms. To overcome the above problem, this paper describes an approach called STNR (suffix tree-based noise resilient) algorithm to detect undemanding and intricate patterns. Undemanding covers symbol and sequence periodicity, and intricate covers segment periodicity. The benefit of STNR algorithm is to detect all the three types of periodicity in a single run. The main contribution in this work is to compare the performance of this algorithm using suffix tree and suffix cactus which act as an underlying data structure. The result shows that the outcome of the STNR algorithm using suffix cactus was more efficient in terms of space and time complexity.