Integrating frequent pattern clustering and branch-and-bound approaches for data partitioning

In this paper, we propose an approach integrating frequent pattern clustering and branch-and-bound algorithms for finding an optimal database partition. First, the Apriori algorithm and cosine similarity are used to determine weighted frequent patterns according to a transaction profile. On the basis of the weighted frequent patterns, we developed two methods for partitioning a database: the candidate method and the optimal method. The optimal method involves using a branch-and-bound algorithm and considering costs in each step of combining attributes until an optimal solution is reached. Furthermore, we refined the optimal method for expediting the execution by reducing the search space. Finally, the experimental results show that the proposed optimal method performs the highest among all examined methods, and the refined method is considerably more efficient than the original method.