Mining High Occupancy Itemsets From Transaction Databases

Abstract

Frequent itemset (FI) mining has been widely studied in data mining over current many years because of its vital role in applications. Nevertheless, the most traditional mining framework in preceding research does not apply correctly for some modern-day applications, which include the travel landscapes recommendation. In 2020, Deng proposed a new algorithm for mining high occupancy itemsets (HO), HEP algorithm (abbreviation for High-Efficient algorithm for mining high occupancy itemsets), where occupancy is the support-based mining structure. It is an efficient algorithm, which helps us to find out all high occupancy itemsets faster than the traditional mining framework. It uses an occupancy-list (OL) structure to store the occupancy and pruning all unpromising itemsets based on upper-bound occupancy (UBO) to mine all HO. However, the HEP algorithm still not optimize in the generating k-itemsets process. In this thesis, we improved the HEP algorithm by two enhancements: add more conditions to prune unqualified itemsets and apply the property of equivalence class to reduce the runtime of the k-itemsets generation process. Finally, we have conducted several experiments on three datasets to prove that the enhancements offer better performance than the original HEP algorithm in terms of execution time and peak memory consumption.