Classification Rule Extraction Based On Interacted Multiple Ant Colonies Optimization Algorithm

The content of database is very rich, which bears a lot of information. Classification is a form of data analysis. Classification analysis as a kind of data mining technology has become an important and hot research areas, and the goal of which is summarizing the general description of each type on the base of existing data classification. In 2002, Rafael S. Parpinelli proposed Classification Algorithm on the base of Ant Colony Optimization.His paper opened up a new applied field for Ant Colony Algorithm. However, precocious and local optimization are the defects of Ant-miner algorithm. For these defects, we proposed Interacted Multiple Ant Colonies Optimization in this paper. The new algorithm makes the following three improvements on the base of the original algorithm:1) Make use of multiple pheromones, and just allows each pheromone build one category. Ant selects category first, and then release the corresponding pheromone.2) Improvement on the state transition rules:instead of the random probability of transition rules we use the Psedo-random probability of transfer rules in the Ant-miner algorithm.3) Use the concept of degree of polymerization to measure the distribution uniformity of rules so as to achieve the purpose of dynamic adjustment path selection strategy. Experimental results show that the new algorithm not only improves prediction accuracy in classification, but also has simpler classification rules than the original algorithm.At the end of the paper, four data of the UCI standard database are used to test the new algorithm. The experimental results show that compared with the original algorithm, the new algorithm not only improves prediction accuracy in classification, but also has simpler classification rules than the original algorithm.