UPFP-growth++: An Efficient Algorithm to Find Periodic-Frequent Patterns in Uncertain Temporal Databases

Palla Likhitha, Rage Veena, Rage Uday Kiran, Koji Zettsu, Masashi Toyoda, Philippe Fournier-Viger

Abstract: Periodic-frequent patterns are an important class of regularities in an uncertain temporal database. However, finding these patterns is computationally challenging due to its enormous search space of \(2^m-1,\) where m represents the number of items (or objects) in a database. Previous studies tried to tackle this problem using some upper-bound constraints. We have observed that these constraints were not tight enough and there exists a possibility to reduce the search space effectively. This paper introduces a new tighter upper-bound constraint, called cutoff expected support (CES), to reduce the search space effectively. This constraint exploits the anti-monotonic nature of the probability (i.e., probability decreases with the increase in the number of items that have to occur simultaneously) to determine whether a superset of a pattern can be a periodic-frequent pattern or not in a database. We also propose an efficient depth-first search algorithm, called Uncertain Periodic-Frequent Pattern-growth++ (UPFP-growth++), to discover the complete set of desired patterns in a database effectively. Emperical results on real-world and synthetic databases demonstrate that CES significantly reduces the search space and UPFP-growth++ is efficient.