Adaptive Minimum Support Threshold for Association Rule Mining

Abstract

In association rule mining (ARM), valuable rules are extracted from frequent itemsets, selecting appropriate minimum support thresholds is essential yet challenging. Arbitrary threshold selection often results in either an overwhelming number of uninteresting rules or the omission of relevant rules. To address this issue, this study introduces an Adaptive Minimum Support (SAd) algorithm designed to dynamically adjust the support threshold based on dataset characteristics, thereby facilitating the discovery of optimal association rules. The SAd algorithm was experimented on three real-world datasets, yielding optimal minimum support thresholds of 0.065, 0.133, and 0.057 respectively. Results demonstrate the algorithm's effectiveness in adapting the support threshold to each dataset's characteristics. By optimizing the threshold, the SAd algorithm enhances the quality of discovered association rules, offering more actionable insights for decision-making.