Analysis And Improvement Of Pheromone Based On Information Gain In Ant Colony Algorithm

Ant Colony algorithm is one of Swarm Intelligence algorithms, and is essentially a meta-heuristic search algorithm. It is inspired by research on the behavior of ant colonies in real world that they are capable of finding the shortest path between food source and their nest. It has been successfully applied to a range of different combinatorial optimization problems and has been applied more and more widely. Data classification is an important branch of Data Mining. With the fast developments of information of technology and Internet, traditional classification methods cannot meet people's need. We need some new data analysis techniques. In terms of the new characteristics of current data, which is widely distributed, dynamic, rough and so on, a number of intelligent data mining technology come into being. Therefore, it is meaningful to research ant colony algorithm for classification rules.In this paper, after in-depth study on Ant Colony algorithm and Ant Miner algorithm, we present an improvement to Ant Miner, which modify the pheromone based on information gain in Ant Colony Algorithm. The innovation is to change the method of initialization of the pheromone. Firstly, information gain is used to sort attributes. According to the results, the higher the value of information gain is, the higher the initial value of the pheromone is. Correspondingly, we change the selection formula and pheromone update formula. Then the algorithm is tested and led to better performance than the original Ant Miner algorithm. It has higher predictive accuracy. And the rule lists are simpler than those discovered by the original one. Finally, we change the method of initialization of the pheromone in distributed database as in a single database. We propose a new algorithm about classification in distributed database based on Ant Colony algorithm, combined with the local and global pheromone.