Research Of Data Classification Means Based On Ant Colony Algorithm

Classification is an important means of data analysis,it has been widely used in such filed as data mining and artificial intelligence.People researched on the problem of data classification in-depth,and produced a variety of classification algorithms,such as:decision trees classification algorithm, Bayesian classification algorithms,and so on.In real life,many problems can be translated into the classification problem,as a result the research on the classification algorithm has a very important practical significance.Ant is a kind of insects that lives in colonies,although their individual acts are simple,but when they act as a whole community,they are capable of solving complex problems in their daily lives,through mutual cooperation, and showing a high degree of self-organization.Inspired by the ants' foraging behavior,M.Dorigo,a Italian scholar,proposed a new type of simulation evolutionary algorithm - ant colony optimization algorithm(ACO algorithm). Since the algorithm has been proposed,its applications have been rapidly expanding,and a breakthrough has been achieved on its hardware, meanwhile it has shown an unprecedented vigor in the improvement of model and the combination with other algorithms.Although,there are a lot of researches on the methods of data classification and ACO algorithm,but applying the ACO algorithm to the field of data mining is less.In 2002,Parepinelli,a British scholar,and his colleagues applied the ACO algorithm to the data mining field,and proposed an new algorithm for solving the classification problem-Ant-Miner algorithm,the algorithm is the first model of classification algorithm based on ACO algorithm.As Ant-Miner algorithm is robust and shows a great potential in resolving the problem of large-scale data classification,and has achieved good results.Subsequently,many scholars improved the Ant-Miner algorithm.For example:in China,Liu Bo,a Professor of Jinan University, proposed the Ant-Miner2 and the Ant-Miner3 algorithm,Zi-Qiang Wang proposed the ACO-Miner algorithm;in other countries,James Smaldon,a British scholar,proposed the Unordered Rule Set Ant-Miner algorithm,and SO on.The paper studies the data classification algorithms,ACO algorithm and Ant-Miner algorithm systemically,and improved the Ant-Miner algorithm from the following two aspects:(1) In order to avoid the Ant-Miner algorithm to emerge the premature stagnation,it proposes an Ant-Miner algorithm with immune feature.The algorithm introduce three kinds of operators to the original Ant-Miner algorithm,thereby enhances the search capabilities of the algorithm. Experimental results show that the improved algorithm has a significant improvement in accuracy rate than the original algorithms.(2) Some shortcomings exist in the original strategy of choice for the conditions,we propose a new strategy of choice of conditions- dual conditions choose strategy.We choose condition should not only consider the value of probability of transfer function but also compare the number of data that be covered by the condition.Applying the new strategy to the original Ant-Miner algorithm,we have an improved Ant-Miner algorithm Ant-Miner Algorithm based on Dual Condition Choose Strategy. Experimental results show that the improved algorithm is superior to the original algorithm not only in the classification accuracy but also in the run-time of the algorithm.