Mining weighted frequent sequences in uncertain databases

Frequent pattern mining has become very useful and interesting to researchers due to its high applicability. Different real-life databases (e.g., sensor network, medical diagnosis data) are uncertain in their nature. Many algorithms have been developed to mine the frequent uncertain patterns based on expected support values. Nonetheless, those are circumscribed to find the frequent patterns by using some filtering constraints. Moreover, it is challenging to find the actual interesting patterns as different patterns carry different importance. In this work, a new framework is proposed to mine sequences in uncertain databases satisfying both weight and support constraints. Subsequently, an efficient algorithm (uWSequence) is developed to discover the uncertain weighted sequences. In addition, the pruning measures iMaxPr, and expSupport(top) play a vital role to make uWSequence remarkably time-efficient, by filtering out the unfavorable patterns in early stages. The applicability of this proposed framework is shown to solve various problems (e.g., weather forecasting, sensor-based event findings). To our knowledge, ours is the first work on weighted sequences in uncertain databases. Extensive performance analysis confirms the efficiency of the proposed algorithm as well as the superiority over the existing algorithms. (C) 2018 Elsevier Inc. All rights reserved.