AN EFFICIENT MODEL FOR MINING PRECISE QUANTITATIVE ASSOCIATION RULES WITH MULTIPLE MINIMUM SUPPORTS

Association rule mining deals with the correlations of items in a transaction. However, this approach raises two problems. First, each item is set with a unified minimum support which cannot be applied in actual applications. The reason is that the frequencies at which items are bought differently. In a shopping case, some items are bought frequently but others are seldom bought because of demand or price. Therefore, setting a unified higher threshold value makes it difficult to find rare items with higher prices, while a unified lower threshold value might lead to a combinational explosion problem. Second, traditional association rules lack quantity-related information and cannot reveal the quantities of different items in an association rule. Therefore, this study proposes quantity-related association rules adopting multiple minimum supports to address these two problems. An efficient algorithm is developed based on a divide-and-conquer idea to find precise quantitative association rules with multiple minimum supports in bag databases. Experiments show the algorithm's computational efficiency and scalability. Our research model can contribute to applications in more realistic circumstances in cases where items occur at various frequencies and have different quantity-related information.