Parallel high average-utility itemset mining using better search space division approach

Abstract

Since the last decade, High Utility Itemset (HUI) mining has emerged as a popular pattern mining approach. HUI mining discovers a set of itemset with their profit more than a user defined profit threshold. High Average-Utility Itemset (HAUI) mining is an improvement over HUI mining that involves the length of items to refine the patterns and keep a fair mining process. In the era of big data, traditional HAUI mining algorithms are not suitable to process large transaction dataset on standalone system due to limitation of processing resources. Therefore, several distributed frameworks have been developed to process big data on cluster of commodity hardwares. This paper presents a parallel version of the traditional HAUI-Miner algorithm and names it as Parallel High-Average Utility Itemset Miner (PHAUIM). PHAUIM is a Spark-based distributed algorithm which splits the dataset into multiple chunks and distributes on cluster nodes to process each data chunk in parallel. In addition, an improved approach for search space division is developed. Proposed search space division technique fairly assigns the workload to each node and upgrades the performance. Comprehensive experiments have been performed to measure the performance of PHAUIM in terms of speedup and data scalability. PHAUIM is also compared with traditional HAUIM.