A fast and resource efficient mining algorithm for discovering frequent patterns in distributed computing environments

•The algorithm is parameter-less and able to determine the appropriate number of computing nodes automatically.•The algorithm requires only 54.3% execution time of PSWS when using 20 nodes.•The algorithm achieves better load balancing as compared with existing methods.•The nodes do not need to exchange any transactions or sub databases with each other.

The advancement of electronic technology enables us to collect logs from various devices. Such logs require detailed analysis in order to be broadly useful. Data mining is a technique that has been widely used to extract hidden information from such data. Data mining is mainly composed of association rules mining, sequent pattern mining, classification and clustering. Association rules mining has attracted significant attention and been successfully applied to various fields. Although the past studies can effectively discover frequent patterns to deduce association rules, execution efficiency is still a critical problem. To speed up execution, many methods using parallel and distributed computing technology have been proposed in recent years. Most of the past studies focused on parallelizing the workload in a high end machine or in distributed computing environments like grid or cloud computing systems; however, very few of them discuss how to efficiently determine the appropriate number of computing nodes, considering execution efficiency and load balancing. An intuition is that execution speed is proportional to the number of computing nodes—that is, more the number of computing nodes, faster is the execution speed. However, this is incorrect for such algorithms because of the inherently algorithmic design. Allocating too many computing nodes can lead to high execution time. In addition to the execution inefficiency, inappropriate resource allocation is a waste of computing power and network bandwidth. At the same time, load cannot be effectively distributed if there are too few nodes allocated. In this paper, we propose a fast, load balancing and resource efficient algorithm named FLR-Mining for discovering frequent patterns in distributed computing systems. FLR-Mining is capable of determining the appropriate number of computing nodes automatically and achieving better load balancing as compared with existing methods. Through empirical evaluation, FLR-Mining is shown to deliver excellent performance in terms of execution efficiency and load balancing.