Effective approaches for mining correlated and low-average-cost patterns

Abstract

High-utility pattern mining is a data-mining research field that focuses on identifying patterns that exceed a specified threshold for utility values. These utility values primarily indicate positive benefits such as profits. However, in actual scenarios, certain applications and demands require the discovery of patterns at a lower cost. Hence, the concept of low-average-cost itemset mining is introduced to evaluate patterns based on their average costs. This is different from high-utility pattern mining, as the cost is minimized. In this study, we focus on extracting patterns with strong correlations and low average costs. Because existing techniques depend on the support metric to devise cost-based pruning strategies, these strategies are ineffective in extracting correlated and low-average-cost patterns. Therefore, we introduce a novel lower bound to efficiently limit the search space for patterns. Experimental analyses are conducted on eight databases to assess the performance of the mining algorithm in terms of execution time, memory usage, pruning effectiveness, and scalability. Additionally, the effects of different pruning techniques and item-sorting orders on the performance of the algorithm are investigated.