A new Dynamic Utility Partitioning Algorithm for High-Utility Itemset Mining in Databases with Unstable Negative Profits

Abstract

High-Utility Itemset Mining (HUIM) is a significant task in data mining, especially in situations with fluctuating negative profits, such as retail discounts and healthcare cost control. Despite the fact that existing algorithms such as EHMIN and EMHUN for mining under hybrid datasets encounter issues such as scalability, execution time, and memory efficiency. To achieve this purpose, this study develops a Dynamic Utility Partitioning (DUP) algorithm that features dynamic item partitioning, Redefined Utility Upper Bound (RUUB) pruning, and an adaptive recursive search strategy. Thus, DUP effectively boosts pruning efficiency and cuts down the computational cost, rendering itself applicable on large-scale datasets. Experimental results on benchmark datasets show that DUP significantly outperforms the state-of-the-art algorithms in execution time by up to 25%, memory usage by nearly 20% compared to EHMIN and EMHUN across benchmark datasets, and candidate reduction. The proposed algorithm can be advantageous in the applications such as retail analytics, healthcare optimization, and supply chain management, where hybrid and unstable utilities are common.