A Three-Scan Mining Algorithm for High On-Shelf Utility Itemsets Guo-Cheng Lan 1 , Tzung-Pei Hong 2, 3 and Vincent S. Tseng 1 1 Department of Computer Science and Information Engineering National Cheng-Kung University Tainan, 701, Taiwan, ROC. rrfoheiay@idb.csie.ncku.edu.tw; tsengsm@mail.ncku.edu.tw 2 Department of Computer Science and Information Engineering National University of Kaohsiung Kaohsiung, 811, Taiwan, ROC. tphong@nuk.edu.tw 3 Department of Computer Science and Engineering National Sun Yat-Sen University Kaohsiung, 804, Taiwan, ROC. ABSTRACT. In this paper, we introduce a new kind of patterns, named high on-shelf utility itemsets, which consider not only individual profit and quantity of each item in a transaction but also on-shelf periods in a database. We have thus proposed a 3-scan mining algorithm to efficiently discover the itemsets. The proposed approach adopts a new pruning strategy and an itemset-generation mechanism to prune redundant candidate itemsets early and to systematically check the itemsets from transactions. Experimental results also show its performance. Keywords: Data mining; Utility mining; High utility itemsets; On-shelf data; Pruning strategy. 1. Introduction. Mining association rules [2] is an important issue in the fields of data mining due to its wide applications. Agrawal et al. [1] first proposed the most well-known algorithm, Apriori, for mining association rules from a transaction database. However, since the association- rule model assumes the same significance or profit for each item, the actual significance of an itemset cannot be easily recognized. Chan et al. thus proposed a new topic, namely utility mining, which discovered high utility itemsets from a transactional database [4]. It considers both individual profit and quantity of each product in a database, thus representing practical utility of an itemset. Several researches about utility mining were proposed in these years, most of which emphasized on how to efficiently find out the high utility itemsets from the databases [6, 10, 11]. Besides, many studies [1, 3, 5, 9] were proposed to dynamically mine association rules. An example for dynamical mining is to find the patterns for on-shelf products. However, a product may be put on shelf and taken off shelf multiple times in a store. If the entire database is considered for mining, the rules discovered may have a bias.