A tree structure frequent pattern mining algorithm based on hybrid search strategy and bitmap

This paper proposes a novel vertical format-based frequent pattern mining algorithm HBMFP. HBMFP adopts a hybrid search strategy of prefix-depth-first and depth-first searches based on the correlative array, which adequately makes use of the advantages of the both searches to effectively reduce yielded candidates as the same time keeps simplicity and lower memory cost. HBMFP uses bitmaps to store the tidsets of itemsets and adopts bitmap projection to compress bitmaps so as to save the time spent in intersecting bitmaps. HBMFP can output the mining results in the sets of itemsets or in tree structure called frequent pattern tree, the later not only takes up less storage space, but also facilitates implementing efficient pattern matching and the visualization of the mining results. Analyses and experiments show that HBMFP has higher mining efficiency, less memory cost, good scalability and operability. Moreover, it can be used in parallel mining.