Article Preview
TopIntroduction
Association rule mining is an important data mining task that aims to extract explicit relationships between the different attributes or items in a large dataset. This mining process can be divided into two essential phases: first, all of the frequent itemsets whose support exceeds a certain threshold are extracted, and second, association rules are generated from the frequent itemsets.
In practice, enumerating all frequent itemsets in a large dataset, especially when dealing with dense datasets or low support threshold values, is very costly (Alves, Rodríguez-Baena, & Aguilar-Ruiz, 2010). Besides, many datasets contain both numerical and categorical attributes that need to introduce more specific techniques to deal with these cases and take into account the different types of attributes and the high-dimensional feature space.
In the recent years, many researchers utilized several meta-heuristic algorithms such as Genetic Algorithms (Kabir, Xu, Kang, & Zhao, 2017) and (Martín, Alcalá-Fdez, Rosete, & Herrera, 2016), Ant Colony Optimization (Manju and C. Kant, 2015) and Particle Swarm Optimization (Yan, Zhao, Lin, & Bai, 2019) to generate association rules sets with different performances by adopting search algorithms to obtain the best quality solutions (rules) in relation to the candidate solutions.
Recently, many hybrid algorithms are proposed for solving optimization problems to take full benefit of the advantages of both GA and PSO algorithm to improve the searchability and increase the exploration of the solution space ((Garg, 2019); (Narang, Patwal, & Garg, 2017) ; (Garg, 2016)). Many others bio-inspired algorithms like Grey Wolf Optimization (GWO), Wind Driven Optimization (WDO) and Whale Optimization Algorithm (WOA) are embedded with principal component analysis (PCA) into the deep neural network (DNN) to choose optimal parameters for training or extract relevant dimensions ((Gadekallu, et al., 2020); (Gadekallu, et al., 2020); (Iwendi, et al., 2020)).
It is worth noting that there is an overlap between works in this field and other areas. For instance, (Lakshman, Kaluri, Gadekallu, Nagaraja, & Subramanian, 2016) where the authors proposed an enhanced algorithm for discovering the most recurrently occurring patterns in biological sequences.
However, according to the NFL Theorem (No Free Lunch) (Wolpert & Macready, 1995), any meta-heuristic algorithm performs only as well as the knowledge concerning the cost function; hence it cannot be capable of dealing with all optimization problems, which would provide perspectives for proposing a new algorithm or improving existing algorithms.
Crow search algorithm (CSA) is one of the recently developed meta-heuristic algorithms successfully used to solve continuous problems by returning interesting results (Askarzadeh, 2016). Due to its simplicity and efficiency, CSA has been used to solve different problems such as feature selection problem (Gupta, et al., 2018), image segmentation (Oliva, et al., 2017), diagnosis of diseases (Gupta, Sundaram, Khanna, Ella Hassanien, & Albuquerque, 2018), electromagnetic optimization (dos Santos Coelho, Richter, Mariani, & Askarzadeh, 2016), economic load dispatch problem (Mohammadi & Abdi, 2018), and Scheduling Problems (Huang, Girsang, Wu, & Chuang, 2019).