Parallel implementation of FP growth algorithm on XML data using multiple GPU

Abstract

The FP Growth algorithm is inherently faster than Apriori as it has less number of combinations to be considered. However, the gap here is that the tree building task is a strenuous process in terms of time and memory. Several attempts have been made to improvise the algorithm. In this paper, a model is proposed to implement a parallel FP Growth algorithm that makes use of the elimination process employed by FP Growth algorithm without generating the actual tree (or multiple smaller trees). This not only improves performance of the algorithm but also results in more efficient memory usage. The proposed algorithm Accelerated Frequent Itemset Mining (AFIM) makes use of multiple Graphics Processing Unit (GPU) system.