The Frequent Pattern List: Another Framework for Mining Frequent Patterns

The mining of frequent patterns (or frequent itemsets) plays an essential role in many tasks of data mining. One major methodology for mining frequent patterns is the Apriori-based approach, which is computationally costly because many candidate itemsets have to be generated and verified. More recently, another approach using the Frequent-Pattern Tree (FP-tree) have been suggested to avoid the generation of candidate itemsets, but at the cost of working with more complex data structures. In this paper, we propose a simpler and more efficient data structure for representing the databases --- the Frequent Pattern List (FPL). The FPL is able to partition both the search space and the solution space so that a divide-and-conquer approach can be applied in mining frequent patterns. With simple operations performed on FPL, frequent patterns can be easily discovered. For a comparative study, we also elaborate the essential differences between FPL and FP-tree in memory requirement, the number of recursive calls, and run time. Experimental results show that our method has satisfactory performances in all these respects. At the end of this paper, we also explore the possible extensions of the Frequent Pattern List in mining dense and large databases.