Search

Classification rules mining is one of the most studied data mining problems. In this article, chemistry based Artificial Chemical Reaction Optimization Algorithm (ACROA) has been for the first time used for mining of generalized classification rules, which is a complex and no well researched generalized variant of classification rules mining where there is more than one goal attribute to be predicted. Furthermore, interestingness measure has been added by performing the adaptations to the algorithm in order to make the rules mined by the algorithm not only accurate and comprehensible but also interesting, surprising, and unexpected. Different rule sets within different databases satisfying different objectives have been flexibly mined by adapting the representation scheme and objective function. Performance of ACROA in classification rules mining within different real public databases have been compared that of genetic algorithm, particle swarm optimization algorithm, and ant colony optimization algorithm. It has shown that performance of ACROA within this special task of data mining is promising. ACROA can be an efficient solution method for different data mining tasks such as association rules mining, clustering rules mining, sequential pattern mining, and etc. [ABSTRACT FROM AUTHOR]

Optimization is the process of finding the best solution of a problem. There are many optimization algorithms proposed for optimization problems. The metaheuristic algorithm can give solutions close to optimum in an acceptable period of time for large-scaled optimization problems. The metaheuristic optimization algorithms are evaluated in seven different groups which are biology-based, physics-based, swarm-based, social-based, music-based, sport-based, and chemistry-based. The swarm-based optimization algorithms have been developed by observing the behaviors of creatures e.g. birds, fishes, cats, bees etc. Data mining is the process of discovery of meaningful and useful data within huge databases. Classification rules mining is one of the most commonly studied data mining problems and with methods for this problem, users can easily understand the rules extracted from databases. In this work, one of the most recent swarm based optimization algorithm, Cat Swarm Optimization (CSO), has been firstly used for classification rules mining within databases composed of numerical or mixed data types automatically. There is not any preprocess for finding true ranges for appropriate attributes of the rules, this has been automatically done by CSO. Furthermore, the used objective function is very flexible and many different objectives can easily be added to. For this purpose, four numerical databases obtained from UCI data warehouse have been used and accurate and comprehensible classification rules have been mined. The results have been compared with the results obtained from Weka program. Although CSO has not been embedded with any improvement and has firstly implemented in this research area, the obtained results seem promising. [ABSTRACT FROM AUTHOR]