A Time-Position Join Method for Periodicity Mining in Time Series Databases

Periodicity mining is used for predicting trends intime series data. There are many applications data includingtemperature, stock prices depicted in the financial market, gene expression data analysis, etc. In general, there are threetypes of periodic patterns which can be detected in the timeseries data: (1) symbol periodicity, (2) sequence periodicityor partial periodic patterns, and (3) segment or full-cycleperiodicity. Rasheed et al. have proposed a two-phasesapproach to periodicity mining. In the first phase, they usethe suffix tree to produce candidate period patterns of allthree types of periodicity in a single run. However, we findthat those suffix-tree-related data structures are stillinefficient in generating candidates of period patterns. Therefore, in this paper, we use the following method forperiodicity mining in time series databases. On the design ofPhase 1 for generation of candidate patterns, we present ourtime-position join method. From the simulation results, weshow that our method is more efficient than their algorithm.