Classification Rule Data Mining Based On PSO

Particle Swarm Optimization (PSO) is a new optimization technique in the artificial intelligence field. Since 1995, it has been gradually applied in the field of optimization and data mining.In this paper, we applied the discrete PSO in the Pittsburgh approach to build a PSO-based classifier. We also propose the concept of the rule mask and combine the rule deletion operator and the minimum description length-based fitness function to make the particle's length variable.In this paper, we also apply the USD algorithm to make the classifier deal with the discrete and continuous attributes simultaneously. Four datasets obtained from the UCI were tested in our experiment, and the results are compared with that of the J48 and Entropy algorithms.The experimental results show that the classification performance of our method excels that of the J48. The proposed method also has the merits of variable number of rules and can simutaniously deal with continuous and discrete attributes.During nearly 10 years of development of particle swarm optimization algorithm, various research issues are presented, and show the growing trend. However, compared with genetic algorithm, technology of relatively long evolution, particle swarm optimization algorithm in data classification problem domain is still not mature, and there still exist many breakthrough points that can be further discussed, therefore, this research attempts to propose some subjects that are still being studied and leave for researchers in future to study and discusses.