Utilizing Bat Algorithm to Optimize Membership Functions for Fuzzy Association Rules Mining

• Published in: Database and Expert Systems Applications Vol. 10438; pp. 496 - 504
• Main Authors: Song, Anping, Song, Jiaxin, Ding, Xuehai, Xu, Guoliang, Chen, Jianjiao
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2017; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Bat algorithm; Fuzzy association rules; Membership functions
• ISBN: 331964467X; 9783319644677
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-319-64468-4_37

In numerous studies on fuzzy association rules mining, membership functions are usually provided by experts. It is unrealistic to predefine appropriate membership functions for every different dataset in real-world applications. In order to solve the problem, metaheuristic algorithms are applied to the membership functions optimization. As a popular metaheuristic method, bat algorithm has been successfully applied to many optimization problems. Thus a novel fuzzy decimal bat algorithm for association rules mining is proposed to automatically extract membership functions from quantitative data. This algorithm has enhanced local and global search capacity. In addition, a new fitness function is proposed to evaluate membership functions. The function takes more factors into account, thus can assess the number of obtained association rules more accurately. Proposed algorithm is compared with several commonly used metaheuristic methods. Experimental results show that the proposed algorithm has better performance, and the new fitness function can evaluate the quality of membership functions more reasonably.