Mining Frequent Patterns Based on Hybrid Projection

Frequent pattern mining is a fundamental data mining problem for which algorithms still suffer from inefficiencies because of the inherent complexities The new algorithm HP presented in this paper discovers frequent patterns by employing hybrid projections of datasets to grow a frequent pattern tree The basic idea is that any dataset cannot be simply classified as dense or sparse one, so the mining algorithm should dynamically adjust its frequent pattern tree search strategies, representations of transaction subsets, projection methods according to features of the local subsets Also proposed in HP are the tree based pseudo projection and array based unfiltered projection that resolves the contradiction between time complexity and space complexity Comparative experiments show that HP is one to three orders of magnitude more efficient than Apriori, FP Growth and H Mine, but also more scalable than other algorithms