Distributed Frequent Itemset Mining with Bitwise Method and Using the Gossip-Based Protocol
Hoda Rafieipour, Azadeh Abdollah Zadeh, Mehrdad Mirzaei · 2020
Nowadays, distributed systems are prevalent and practical in network environments. In distributed systems, pattern recognition help to extract information from network nodes. Meanwhile, data mining in such systems needs resource consideration in terms of storage and computational time. The primary requirement of these systems is a scalable mechanism to distribute the tasks on several databases. Moreover, to do a centralized process, relocating data from all nodes or partial nodes to a central node has confidential risks and traffic overhead. Therefore, distributed data mining in distributed environments needs systematic and structural techniques. In this paper, we propose a new algorithm to extract frequent itemsets in Wireless Sensor Networks. Through this algorithm, nodes frequent local itemsets are obtained with a Bitwise approach, and nodes are classified into clusters by using the Low Energy-Adaptive Clustering Hierarchy (LEACH) algorithm. Connecting the head cluster is performed by a Gossip-based protocol to achieve the values of global support, and it finally resulted in the extraction of frequent itemsets. The proposed algorithm has been simulated in various scenarios using Java software, and algorithm efficiency is evaluated in terms of execution time and average accuracy. Our algorithm is compared with a Gossip-based algorithm, and then some improvements in execution time have been presented.