Algorithms For Mining Horizontally Weighted Association Rules

Association rule mining is an important problem in data mining. Many association rule mining methods make no difference in item's importance. However, in some real applications, users often attach different importance to different items, so items should have different weights. The main contributions of this paper are as follows: We present three different algorithms, Direct algorithm, Reordered algorithm and Fp-tree based algorithm, for horizontally weighted association rule mining problem. Theoretical analysis and experimental results show that the Fp-tree based algorithm is the most efficient one and the Reordered algorithm generates the least candidate sets. The Reorder algorithm is more efficient than the New_Apriori algorithm when mining large databases. Furthermore, we discuss how these algorithms can be used in mining association rules with item constraints. In the traditional association rule mining algorithms, all items belong to the same class. In practice, however, items are often divided into different classes. And the more classes an itemset contains, the more important the itemset could be. Therefore, we deal with a new category of association rule mining problem, namely, the Item_Class_Based association rule mining. For solving this problem, two items reordered methods are proposed, namely, reorder by weight and reorder by class. And two algorithms are presented with these methods, namely, Reorder Apriori algorithm and Reorder Fp-tree based algorithm. Experimental results show that the Reorder Fp-tree based algorithm is more efficient than the other, and the algorithm implemented based on reorder by class is more efficient than the algorithm based on the other reorder method. The OOA mining intends to mine all association rules related to a given objective based on support, confidence and utility. We present a new approach to OOA mining using Disjunctive-free pattern algorithm. Our experimental results show that the algorithm is much more efficient than the algorithm based on Apriori when there is relatively strong dependency among data.