Stability of Low-Cost High-Utility Patterns Under Uncertain Cost Assignments

Abstract

Low-Cost High-Utility Itemset Mining (LCHUIM) is a recent extension of utility-based pattern mining that aims to identify itemsets with high utility and low associated cost. In many real-world applications, especially in domains like education or healthcare, explicit cost information may be unavailable or difficult to measure. This study investigates the stability of LCHUIM under uncertain cost settings by applying it to a real-world educational dataset where cost values are not explicitly provided. We generate three different cost assignment strategies using interpretable mappings and evaluate the impact of cost differences under various utility and support thresholds. Experimental results show that under stricter thresholds, LCHUIM yields highly stable results. As thresholds are relaxed, more patterns emerge, and the sensitivity to cost differences increases. Nevertheless, a considerable number of patterns remain consistent, indicating that LCHUIM is capable of producing reliable insights even when cost values are estimated. This work contributes to understanding the robustness of utility-based pattern mining in behavior-driven domains with incomplete or estimated cost values. © 2025 IEEE.