MINING MOST GENERALIZATION ASSOCIATION RULES BASED ON FREQUENT CLOSED ITEMSET

摘要

Association rule mining plays an important role in knowledge discovery and data mining. The rules obtained by some previous works based on support and confidence measures might be redundant to a certain degree. This paper thus proposes the concept of most generalization association rules (MGARs), which are more compact than the three previous rule types that include traditional association rules, non-redundant association rules and minimal non-redundant association rules. Some theorems relating to the properties of MGARs are derived as well, and an algorithm based on the theorems for effectively pruning unpromising rules early is then proposed. Hash tables are used to check whether the generated rules are redundant or not. Experimental results show that the number of MGARs generated from a database is much smaller than that of non-redundant association rules and that of minimal non-redundant association rules.