Research Of Applying Ant Colony Optimization In Data Mining

Social insects have high swarm intelligence. Inspired by this mechanism, Ant colony algorithm (ACA) has emerged recently as an evolutionary algorithm or a new meta-heuristic algorithm and has been widely applied to solving NP-hard problems in combinatorial optimization. In this paper, combined with ACO, we mainly make research application on each task in data mining(DM). We not only introduce the advantage of ACO but also analyze kinds of current data mining algorithms at home and abroad. Accordingly, we advance some ACO algorithms for solving clustering problem, classification problem and association rule discovery in data mining.As for clustering problem aspect, we analyze the current classical clustering algorithms and several ant clustering models based on the behavior of piling ants'corpses. In the models of BM and LF, ants act as porters to transit the data to the proper places to form clusters through the behaviors of"pick up"and"drop down". We advance two kinds of adaptive clustering algorithms through research. Firstly, we design the ant active/sleeping clustering model(ASCM). ASCM makes improvement on BM and LF where each ant represents a datum. Ants make the data into clusters through the active and sleeping states. Our algorithm is efficient with better results. Secondly, by analyzing the characteristic of ants system, we make ants connect the similar data. We advance an adaptive ant clustering algorithm based on digraph(A3CD) which is more direct and easier. We make pheromone left by ants in the seeking process as the standard of clustering and design the initial pheromone digraph to strengthen the positive feedback of ants. The weights of the digraph are updated continually based on the adaptive parameters which make the algorithm faster. At last, strong connected components are extracted as clusters under a certain threshold. Compared with the current clustering algorithms, our experimental results show that our algorithm is self-organizing , adaptive, efficient and has better clustering quality.As for the problem of classification rule discovery, we also introduce some current classification methods, such as C4.5 which is based on the classical decision tree, and...