A Multithread, Novel Pattern Based Algorithm for Finding Frequent Patterns With Jagged Array and Vertical Data Format

Abstract

Frequent pattern mining is essential for discovering hidden items from a database with more than a prescribed threshold. Knowing frequent patterns helps us to determine the relationship between the items. Many researchers narrated novel algorithms for sequential frequent itemset mining using a single thread, but still, there is a need for time, memory efficient and scalable one. Therefore, the research study proposed an approach for finding frequent patterns, namely TB-NPF-VDF (Thread Based, Novel Pattern Formations with Vertical Data Format), which uses a new way of generating candidate items to minimize the time. Also, it employs a multithread concept and runs several threads simultaneously, one for each frequent 1-itemset to generate the remaining frequent itemsets for that item. Further, it also employs a jagged array to store the frequent patterns to reduce the memory requirement. The research work has been implemented and tested using four real-time datasets. Further, it has been compared with Matrix-Apriori, VDF and NPF-VDF (without multithread), and the experimental results reveal that TB-NPF-VDF outperforms significantly in terms of runtime and storage.