Association Rule Mining Based on Hybrid Whale Optimization Algorithm

Introduction

Data mining is a general and persuasive technique for extracting valuable knowledge from data sources (Telikani et al., 2020). Association Rule Mining (ARM) is one of the most ordinary and crucial tasks in data mining, which can find the association between the data by mining the frequent itemsets (Baró et al., 2020). ARM has high practical value in real life because it contributes to summarizing laws from big data, which has been extensively applied in the fields of healthcare, recommender systems, market analysis, and transportation (Das et al., 2021). For example, Anand Hareendran and Vinod Chandra (2017) extracted the features and their relationships from liver transplantation data using ARM and designed high precise mining model. Viktoratos et al. (2018) proposed a novel approach by combining community-based knowledge with ARM to relieve the cold start problem in recommender systems. Wen et al. (2019) designed a hybrid temporal ARM to predict traffic congestion, and Hong et al. (2020) employed ARM to discover the contributory crash-risk factors of vehicle-involved crashes in the transportation field. The primary purpose of ARM is to find rules that satisfy the predefined minimum support and confidence levels from a given database. Apriori algorithm and FP-growth algorithm are the most typical ARM algorithms. The principle of these algorithms requires scanning the entire data, selecting frequent itemsets by meeting the minimum support threshold, and then obtaining association rules based on the minimum confidence level (Chiclana et al., 2018). Traditional algorithms have low efficiency and high memory overhead for data processing in massive data sets. One of the most effective approaches is combining the optimizer technique with ARM, and previous studies showed that the meta-heuristic method based on population was an effective way to solve the optimization problem (Shu et al., 2022).

Therefore, many swarm-inspired algorithms have been proposed for mining a good subset of association rules. For example, Sarath and Ravi (2013) proposed a binary particle swarm optimization (BPSO) algorithm to generate association rules by constructing a combinatorial global optimizer problem. Kuo et al. (2019) used Pareto-based PSO to optimize the goals, including comprehensibility, confidence, and surprisingness to discover valuable and interesting association rules from numerical valued data sets. Jyoti and Sharma (2020) focused on applying rule mining techniques based on ant colony optimization (ACO) in data classification. Mlakar et al. (2017) designed a modified single-objective binary cuckoo search (MBCS-ARM) that included novel representations of individuals, which helped deal with high dimension problems with an increasing number of attributes. Heraguemi et al. (2015) proposed a bat-based algorithm for ARM (BAT-ARM) and then designed a novel multi-objective bat algorithm for ARM (MOB-ARM) and applied it (Heraguemi et al., 2018). However, most relevant algorithms for controlling association rules are often computationally expensive and possibly generate many irrelevant rules (Kumari et al., 2019).