Mining Top-K Cross-Level High Utility Itemsets

Abstract

High utility itemset (HUI) mining extracts frequent itemsets with high utility values from transactional databases. Traditional algorithms have limitations in detecting relationships between items and categories across multiple levels of a taxonomy-based database. Multi-level and cross-level algorithms have been proposed to address this issue while top-k algorithms find the top-k HUIs with the highest utility values. FEACP and TKC algorithms were proposed for HUI mining with high efficiency. However, they suffer from scalability and efficiency issues when dealing with large datasets. To overcome these limitations, we propose a new algorithm called TKC-E (Efficient Top-K Cross-level high utility itemset miner), which combines the strengths of FEACP and TKC while applying efficient strategies to identify cross-level HUIs in taxonomy-based databases, resulting in significantly improved scalability and efficiency. Experimental results show that TKC-E outperforms TKC in terms of processing speed and memory usage, with up to 4 times memory and 60 times runtime improvements on sparse and dense datasets, respectively