Yildirim, I.2026-03-262026-03-262024979833153149210.1109/IDAP64064.2024.107109262-s2.0-85207958151https://doi.org/10.1109/IDAP64064.2024.10710926https://hdl.handle.net/20.500.14901/3615Mining high-utility itemsets is a popular area within itemset mining, aiming to discover itemsets that provide high utility (profit) in a given dataset. However, it does not consider the investment costs required to obtain these profits. To address this, the problem of high-efficiency itemset mining (HEIM) was recently introduced to discover high-efficiency itemsets based on a given minimum efficiency threshold. In HEIM, the efficiency of an itemset is defined as its utility derived from transactions containing it, divided by its total investment. However, the efficiency of an itemset may increase with its length since the length of itemsets is not considered in the efficiency calculation. Therefore, it may be unfair to compare the efficiency of itemsets of different lengths using the same threshold value. In this study, the problem of high average-efficiency itemset mining (HAEIM) is introduced as an extension of HEIM problem. To evaluate the importance of itemsets with a fairer measurement, HAEIM takes into account not only the utility and investment information of itemsets but also their lengths. An algorithm named HAEIMiner is developed for finding high average-efficiency itemsets. Additionally, an upper-bound model is introduced to improve the mining process by pruning the search space. The results of the experiments conducted to demonstrate the effectiveness of HAEIMiner show that the number of high average-efficiency itemsets is fewer than the number of high-efficiency itemsets under the same threshold. This indicates that the HAEIM problem can be addressed with lower efficiency threshold values compared to the HEIM problem. © 2024 IEEE.eninfo:eu-repo/semantics/closedAccessAverage-EfficiencyData MiningEfficiencyInvestmentItemset MiningConference Object