A two-phase approach for mining weighted partial periodic patterns

Abstract

Partial periodic pattern mining has recently become an important issue in the field of data mining due to its wide applications in many businesses. A partial periodic pattern considers part of but not all the events within a specific period length, repeating with high frequency in an event sequence. Traditional partial periodic pattern mining, however, only considered the frequencies of patterns, but did not consider events that might have different importance. The study thus proposes a weighted partial periodic patterns mining algorithm to resolve this problem. To increase the efficiency, the two-phase upper-bound weighted model based on segmental maximum weights is adopted to prune unimportant candidates in early stage. Then the weighted partial periodic patterns are discovered from the candidate patterns. Finally, the experimental results on synthetic datasets and a real oil dataset show that the weighted partial periodic pattern mining is more practical to assist users for decision making.