The Combinations of Frequent Pattern Tree and Candidate Generation for Mining Frequent Patterns

Many approaches have been proposed for mining frequent pattern. However, either the search space or memory space is huge, such that the performance for the previous approach degrades when the database is massive or the threshold for mining frequent patterns is low. In this paper, we propose an algorithm for mining frequent patterns. Our algorithm only needs to construct a FP-tree and traverse each subtree of the FP-tree to generate all the frequent patterns for an item without constructing any other subtrees. After traversing a subtree for an item, our approach merges and removes the subtree to reduce the FP-tree smaller and smaller. By this way, only a subtree of the reduced FP-tree needs to be traversed to generate frequent patterns for the other item. Since there is no extra trees constructed and the frequent patterns generated for an item only need to traverse a subtree, our approach is much more efficient than FP-Growth algorithm. The experimental results also show that our approach outperforms FP-Growth algorithm.