Efficient Mining of Partial Periodic Patterns with Individual Event Support Thresholds Using Minimum Constraints
Partial periodic patterns are commonly seen in real-life applications and provide useful prediction with uncertainty. Most previous approaches have set a single minimum support threshold for all events to assume they have similar frequencies which is not practical for real-world applications. Instead of setting a single minimum support threshold for all events, Chen et al. proposed an FP-tree-like algorithm to allow multiple minimum supports for reflecting the natures of the events. However, such a tree-based algorithm encountered an efficiency problem while period length is long or event sequential orders in period segments are varied. Under the circumstance, many tree branches are created and much execution time is spent to find partial periodic patterns. In this paper, we thus propose a projection-based algorithm which examines only prefix subsequences and projects only corresponding postfix subsequences with multiple minimum supports to quickly find the partial periodic patterns in a recursive process. Experiments on both synthetic and real-life datasets show that the proposed algorithm is more efficient than the previous one.