Research And Application Of Association Rule Learning On Ant Colony Optimization

With the fast development of the database technology and the wide application of database management system, massive quantities of data have been collected and stored. There may be valuable information behind of the data. People want to take high-level analyses of the data, so as to take good use of them. Data mining technique is the achievement of long term research in the database technology. It is a technique that aims to analyze and understand the data and reveal the knowledge hidden in the data. Data mining technique has been one of major mains for modern information processing.Association rule mining is to discover the unknown, hidden and interesting relations among the data in one data set. It is an important research area of data mining and has extensive application prospect. As an active topic in data mining, many researches about association rule mining mainly focus on the efficiency of the algorisms, ignoring the quality of the rules.Ant Colony Optimization (ACO) algorithm is a kind of intelligent algorithm that simulate the foraging behavior of ant swarm. Taking use of positive feedback scheme, ACO algorithm has the features of robustness, distributed computation and simplicity of combining with other techniques, and it has displayed excellent performance and huge potential of application in many fields such as solving difficult combinational optimization problems. Applying ACO algorithm to association rule mining is a new researching direction, which has been applied in classification decision, cluster analysis and rule discovery.In this thesis, we propose a new method for mining association rules using Ant Colony Optimization (ACO) algorithm. The ACO algorithm and classical Apriori algorithm for association rule mining are combined to find the association rule for frequent items. Firstly, we constructs a complete graph of frequent items based on the frequent sets, and set the support degree of the frequent items as the arc of the complete graph. Then the complete graph can be viewed as the solution space of the problem. The degrees of support among frequent items are taken as heuristic factors for routing. Due to great affect to the pheromone on the arcs by the support, the supports of the frequent items on the vertex are taken into account. After certain iterations, the pheromone on the arc indicates the relations between items. It is the way that ACO shows the association rules. Lastly, the association rules are generated from the graph according to the pheromone on the arcs. Certainly, the prune step should be used to generate rules.Taking the official technique statistics of normal games from 1974 to 2006 of America Baseball League as the experiment data, this dissertation verifies the validity of the proposed algorithm. The results of the proposed algorithm are close to the actual experiences.