Abstract: In FP-growth algorithm, two traversals of FP-tree are needed for constructing the new conditional FP-tree. A novel FP-array technique was presented that greatly reduced the need to traverse FP-trees. A improved FP-growth algorithm was presented which uses the FP-tree data structure in combination with the FP-array technique efficiently. Experimental results show that the new algorithm works especially well for sparse datasets.

Key words: FP-growth algorithm, FP-tree, Sparse databases, FP-array