The Research On The Parallel Algorithm Of Association Rule Mining

The discovering association rules is an important research in data mining, which may find certain specific rules in the magnanimous data. Frequent patterns mining plays a crucial role in association rule mining, which dominates the time complexity and application range of the discovery algorithm. Because of the time-consuming in mining frequent patterns, so it is important to design high performance parallel mining algorithm.This thesis analyzes the performance of typical serial algorithms and parallel algorithms, and introduces their virtues and disadvantages. For the shortage of the traditional algorithm, two parallel algorithms to mine frequent patterns are proposed. The major work is as follow:Firstly, in order to overcome generating large number of conditional pattern bases and high communication, a new parallel algorithm named PBFI-Miner for mining frequent patterns is presented. This algorithm uses the bit objects to express data and to improve the FP-tree, and uses parting strategy to achieve near optimal balancing between processors. The processors communicate with bit object groups so that efficiency of the algorithm is made better. Experimental result verifies the efficiency of the PBFI-Miner.Secondly, for purpose of adapting the characteristics of data streams, a parallel algorithm based on FP-DS for mining frequent patterns from data stream is proposed. PFP-DS algorithm uses the master-slave mode design, communication occurs only between master and slave processor. The host processor is responsible for updating the local FP-DS and mining frequent patterns. The minor processors are responsible for counting the 1-itemset and constructing local FP-DS,which is stored frequent patterns compressed. The experimental results indicate that PFP-DS algorithm performs efficiently.