International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 01 | Jan 2025
p-ISSN: 2395-0072
www.irjet.net
DU-HUIM: a novel dynamic utility high-utility itemset miner algorithm for dynamic utility variations in big data Arkan A. Ghaib Department of Information Technologies, Management Technical College, Southern Technical University, Basrah, 61004, Iraq. ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - High-Utility Itemset Mining (HUIM) has recently
be a product of internal utility (e.g. the Transactional volume) and external utility (e.g. profit per unit) High-Utility Itemsets (HUIs) [5] are the itemsets whose utility is higher than a pre-defined threshold. While FIM can utilize antimonotonicity properties for pruning, this does not hold for HUIM making the mining beyond costly. Techniques to approximate utility, like Transaction-Weighted Utility (TWU) [6], have been developed to alleviate this search space problem. To improve the efficiency and scalability of HUIM, several algorithms have been presented. Absolute HighUtility Itemset Miner (AHUIM) [7] presented a parallel approach for mining huge datasets effectively using pruning strategies such as Absolute Subtree Utility (ASU) and Absolute Local Utility (ALU) in order to speed up computations. Recent improvements to the enhanced models like EFIM-Par and PHUI-Miner made use of distributed and parallel computing frameworks to scale even more. EFIMPar proposed efficient usage of memory and designed pruning strategies that can be applied to large-scale datasets [5,8]. To address the issue of discovering frequent patterns on massive data sets, MRFP-Growth is being used. With the large datasets, most of the algorithms for frequent patterns fail to work. [9,10]
emerged as an important data mining technique that focuses on discovering those itemsets that offer high utility. Nonetheless, this is limited in applicability with real-world settings where utility values change dynamically over time, as the vast majority of existing methods either assume static values for items. In order to overcome this limitation, we propose for the first time in a big data context a new algorithm called Dynamic Utility High-Utility Itemset Miner (DU-HUIM), that efficiently manages these dynamic utility variations. DU-HUIM contains a dynamic utility list structure, adaptive pruning strategies, and parallel processing abilities to achieve high-utility itemset mining. On standard datasets, the experimental results show that DU-HUIM exhibits a remarkable improvement over the prior art with respect to running time, scalability, and resistance to changes in utility. Key Words: Big Data, High Utility Itemset, Frequent Itemset Mining, Parallel computing, Dynamic Miner.
1.INTRODUCTION The problem of High-Utility Itemset Mining (HUIM) [1] has been extensively studied and has various applications in different domains, such as retail, e-commerce, and healthcare. It then finds itemsets whose utility values are higher than a specified minimum utility value, where utility indicates the importance of an item, such as value, weight, or other metrics. However, traditional HUIM algorithms assume that utility values are constant over all transactions, an assumption that can be unrealistic in dynamic environments. For example, prices and demand for products can change depending on things like seasons, promotions, or the market environment. Especially, Frequent Itemset Mining (FIM) [2], a canonical problem in data mining, finds frequent patterns from transactional databases. FIM methods like Apriori and FP-Growth [3] focus on the frequency of itemsets and don't consider the utility or significance of items in practical situations. High-Utility Itemset Mining (HUIM) broadens FIM from the search of itemsets yield itemset to the search of itemsets yield high utility which utility is a metric for the significance, profitablity or importance of an itemset in a dataset(Pandi). This has converged HUIM to become of paramount importance for areas like retail analytics, e-commerce, healthcare, picking and replenishment, inventory and supply chain management [4]. In HUIM, utility per item can
© 2025, IRJET
|
Impact Factor value: 8.315
However, most of these HUIM algorithms still assume static utility values for the entire transactional database. This assumption limits these algorithms because in the real world, the utility values can dynamically change and this can be affected by: Seasonal patterns: In the retail sector, prices and user demand vary with seasons or promotional campaigns. Dynamic pricing: In e-commerce, based on algorithms, prices are variable as per supply and demand. Resource-constrained: In healthcare, utility may depend on urgency of allocating limited resources. Static utility assumptions lead to the wrong insights having been made and actionable decision opportunities missing. For example, AHUIM and EFIM-Par can retrieve goodquality patterns from static datasets, but they are not well equipped to address temporal changes in the utility of patterns. In this paper, we fill the gap and propose the Dynamic Utility High-Utility Itemset Miner (DU-HUIM), the first
|
ISO 9001:2008 Certified Journal
|
Page 405