Research On Classification Rules Mining Based On ACO And PSO Algorithm

Evolutionary Algorithms(EAs)is a global optimization method which has wide range of applicability.Having the advantages of self-organizing,adaptive,self-learning,it is able to get rid of constrains of subject properties itself and can effectively resolve complex optimization algorithms issues which the traditional method barely can solve.Particle Swarm Optimization(PSO)and Ant Colony Optimization(ACO)both are new evolutionary algorithms,with the features of parallelism,positive feedback and synergy,having excellent adaptability in solving complex optimization problems,they have been widely caught attention and utilized in many areas.In the field of data mining research,PSO and ACO algorithm regarding as the method of seeking optimal solution,have been getting attention by researchers in this area.In this paper,the application of PSO and ASO in classification rule mining of data mining is researched,the existed issues when applying these two algorithms in classification rule mining is analyzed,and improvements for the original algorithm is proposed.Finally,a hybrid classification algorithm integrating ACO with PSO is proposed.The main contents of this paper include:Firstly,PSO algorithm has an issue of easily finding local optimal solution.Based on the fundamentals of PSO algorithm,a new approach named Dynamic Adjusting Population Strategy is presented.In the process of searching,through adjusting the parameters of the algorithm adaptively to improve the optimization performance of the algorithm,it strengthens the particle swarm's capacity of escaping from local optima.Experiments proofed that the improved PSO algorithm has advantageous for the particles to jump out of the local optimal solution,improves particle swarm's capabilities of global optimization,it can mine out better classification rules when applied in classification rule mining.Secondly,when ACO algorithm used in data mining,the algorithm can be trapped in local optima and make error when selecting the best rule.This paper analyzes the reason that causes the potential error to choosing the best rule when ACO algorithm is applied in classification,and improvements are proposed.The proposed is a new optimal rule replacement strategy with a precondition,when this condition is met,the best rule filtered out will be replaced.Experimental results indicate that comparing to the original ACO algorithm,the improved ACO algorithm has betterclassification accuracy on rule mined by avoiding local optima when selecting the best rule.Finally,in order to ensure the consistency of the data,in the classifier based on PSO algorithm,the continuous data must be discretized or doing map operation in discrete and nominal data before classification.This paper presents a hybrid algorithm based on ACO and PSO,the improved ACO algorithm is used to deal with discrete attributes and nominal attribute,and the improved PSO algorithm is used to deal with continuous attributes.This avoids the above two operations in dataset.Intrusion detection experiments are carried out on the proposed new algorithm based on KDDCUP99,and experiments show that the rules mined by this new algorithm has a good result in classification.