Optimized Closed Frequent High Utility Itemset Mining Using OSR, OWL, and MSU Pruning on Retail Transaction Data

Abstract

This research proposes the optimization of the Frequent Closed High-Utility Itemset Mining (FCHUIM) algorithm for retail transaction datasets using heuristic-based pruning techniques, Observed Support Ratio (OSR), Observed Weighted Lift (OWL), and Modified Subtree Utility (MSU). The algorithm aims to efficiently extract high-value itemsets that are both frequent and economically significant while minimizing redundant patterns through closed itemset mining. A real-world retail dataset from a consumer cooperative, comprising 56,274 transactions and 4,265 unique items, was used in the experiments. The study evaluates the effectiveness of each pruning technique, individually and in combination, across multiple scenarios of minimum support and utility thresholds. Results show that the proposed optimizations reduce the search space by up to 92.5%, significantly lowering execution time and memory usage. Sensitivity analyses reveal that the minimum utility parameter has a stronger impact on computational efficiency than minimum support, while scalability tests confirm the algorithm's ability to handle increasing dataset sizes with linear performance degradation. These findings confirm that the optimized FCHUIM algorithm is suitable for large-scale retail data mining applications, especially in scenarios requiring fast and concise pattern extraction. Future work may explore real-time integration into recommendation systems and adaptive thresholding for dynamic retail environments.