Discovering periodic cluster patterns in event sequence databases

Guisheng Chen, Zhanshan Li

Abstract: Since periodic events are very common everywhere, periodic pattern mining is increasingly more important in today’s data mining domain. However, there is currently no uniform definition of periodic patterns, and all of these definitions are incapable of discovering seasonally prevalent events. In this paper, we first define the periodic event based on the coefficient of variation of the event’s periods in event sequence. Then, in order to discover seasonally prevalent events, we propose a new concept of periodic cluster patterns and design an efficient algorithm named the PCPM(Periodic Cluster Pattern Miner) to mine periodic cluster patterns in event sequence datasets. To illustrate the application of periodic cluster patterns, we propose a new method employed periodic cluster pattern prediction for next basket recommendation, and the method is named PCPP(Periodic Cluster Pattern Predictor). Experiments show that the PCPM is effective for periodic cluster pattern mining and that PCPP has performances close to those of the baseline methods on four real-world transaction datasets. Furthermore, we believe that periodic cluster patterns, as a new concept, will have a wider application in other domains, such as time series prediction, meteorological forecasting, etc.