| The multi-level association rule mining is an important research direction in the field of association rules. Frequent pattern mining plays a time-consuming part in the multi-level association rule mining, which dominates the efficiency of the association rule mining. Therefore, the design of high performance frequent pattern mining algorithm has important significance. This paper presents two algorithms of frequent pattern mining based on the structure of FP-tree which is analysed deeply.A new algorithm called FPIFS is presented, which is based on the structure of FP-tree, it contraposes the defect of low time efficiency in the FP-growth algorithm. The algorithm avoids traveling the same road many times, and adds a prefix path domain in each node domain of FP-tree, which is used to store all precursors of the node. The precursors of most nodes can be found only need to scan several nodes, and it is not required to scan these roads repeatedly while parents of these nodes are being handled, which improves the time efficiency.A multi-level association rule mining algorithm called FP-ML ,which is based on the FP-tree is presented, it contraposes the shortcoming that most of multi-level association rule mining algorithms can only mine association rules on the same level. Firstly, the item is found which is infrequent in the current layer but its parent item is frequent, and it is replaced by parent item. Secondly, new transaction record is mined by FPIFS algorithm. Finally, frequent pattern which belongs to different conceptual level is mined .Comparing the FPIFS algorithm and the FP-ML algorithm with FPIFS algorithm and FP-ML algorithm separately on the T10I4D100K dataset and retail dataset. It shows that FPIFS algorithm is lower than FP-growth algorithm on the time efficiency, and its stability is better than FP-growth algorithm. Meanwhile, FP-ML algorithm is lower than MLAR-FP algorithm on the time efficiency, and its stability is better than MLAR-FP algorithm. |
PDF Full Text Request