A Peer-to-Peer Approach to Parallel Association Rule Mining

Distributed computing based on P2P (peer-to-peer) networks is a technology attainable at a relatively low cost. This enables us to propose a flexible approach based on “Partition” algorithm as an extension of “Apriori” algorithm to efficiently mine association rules by cooperatively partitioning and distributing processes to nodes on a virtually tree-like P2P network topology. The concept of cooperation here means that any internal node contributes to the control of the whole processes. First, we describe the general design of our basic approach and compare it with related techniques. We explain the basic algorithm (without load balancing) implemented as experimental programs in detail. Next, we explain simulation settings and discuss evaluation results, which can validate the effectiveness of our basic approach. Further, we describe and evaluate the algorithm with load balancing as an extension to the basic algorithm.