Negative Itemset Tree for Discovering Rare Patterns with Periodicity from Static Databases

Periodic patterns assist in the discovery of crucial information in the fields, including fraud detection, telecommu-nications, retail marketing, research, and medicine. Numerous methods have been developed for the extraction of periodic frequent patterns. The capacity to find uncommon or unexpected combinations that are missed by periodic frequent pattern mining techniques is the main motivation for Periodic Rare Pattern Mining. If the rare patterns are distributed across the transaction dataset, they are periodic and important. A novel PRPNegTidTreeMiner algorithm is proposed to discover rare patterns with periodicity along with support threshold. An efficient tree structure, NegTidTree, is constructed to capture the complete negative representation of the given static database. NegTidTree, serves the dual purpose of finding support count as well as periodicity information. This tree accelerates the extraction of periodic rare patterns and prevents repeated database scanning. PRPNegTidTreeMiner uses two mining techniques to generate periodic rare patterns. The first mining method, NegTidTreeMiner, employs a top-down approach to find rare patterns with periodicity thresholds which avoids traversal of frequent patterns. In the second technique, NegTidTreeMiner-FLP, the mining efficiency is improvised by avoiding the traver-sal of frequent as well as non-existing patterns. Experiments are carried out by varying support and periodicity components for a variety of datasets. The results show that NegTidTreeMiner performs better than NegTidTreeMinerFLP when the dataset is small and generates a huge number of periodic rare patterns. When the size of the database grows, NegTidTreeMinerFLP continuously outperforms NegTidTreeMiner.