Abstract: Finding maximal frequent itemsets is an important issue in data mining research field．The FPMAX algorithm,which is based on the FP-tree structure,has been proved to be one of the high-performance algorithms on maximal frequent itemsets mining．But for data mining task in dense datasets,FPMAX algorithm will construct a large number of redundant conditional FP-tree．What’s more,if the quantity of frequent itemsets is large,the MFI-tree structure used for subset testing in FPMAX will become quite big,decreasing the efficiency of subset testing in the algorithm．Therefore,this paper proposed the FPMAX-reduce algorithm to overcome those drawbacks of FPMAX．This novel algorithm uses a pruning technique based on the common suffix of transactions and greatly reduces the construction of redundant conditional FP-tree．Besides,when the scale of the newly constructed conditional FP-tree is small,FPMAX-reduce constructs a corresponding conditional MFI-tree,which deletes the redundant information,to improves the efficiency of subset testing in the following recursive calls．Experimental results show that FPMAX-reduce algorithm effectively improves the efficiency of FPMAX and outperforms many existing available algorithms in dense datasets.

Key words: Frequent itemset,Maximal frequent itemset,FP-tree,FPMAX,FP-growth