Bi-directional partitioning approach to frequent item mining

Frequent Itemset Mining (FIM) is one of the classical and well-adopted descriptive approaches in data mining. However classical algorithms of FIM like Apriori method suffer from higher I/O overhead, inability to scale-up for higher items (dimensions) and higher computational time requirements. To overcome the I/O overhead, Partition algorithm is proposed; however, it does not scale-up well for large number of items and computes candidate/frequent itemsets almost m (> 1) times more than Apriori based methods, there by increasing computational time complexity. In this work, we propose a hybrid Bi-directional partitioning approach which computes some frequent itemsets using itemset partitioning and computes residual frequent itemsets using horizontal partitioning like Partition algorithm. The proposed method is proved to be superior, upto a factor of s/⌈log
₂
s⌉ + 1 in terms of I/O overhead over Apriori based methods; it is computationally much better than Partitioning algorithm. The proposed method is scalable to large number of items as it makes use of smaller sub-itemsets for computation of candidate itemsets.