Abstract: There are excessive and messy rules produced by traditional association rule mining, many of which are not relevant to users' interest. Minimal association rules mining algorithm was represented based on the concept of minimal association rules-set and strong dependency between items. Not only avoiding checking whether every non-empty subset of one frectuent itemset can form an association rule,but also simplifying the traditional rules set by deleting those excessivcly complex and reduplicative rules. I}he support and confidence degree of most redundant rules can be derived from minimal association rules set, which achieves a nearly lossless representation of the traditional rules set. The recults based on four benchmark data sets from UCI repository show that the number of rules generated by proposed method is reduced greatly and those rules in rules set arc more briefly without reduplicative information. This provides a better way to find minimal association rules.

Key words: Minimal association rules,Itemset dependency,Redundant rules