Ant Clustering Algorithm With K-harmonic Means Clustering

Data mining is the process that obtaining information and knowledge from a lot of imperfect noisy random data. Clustering analysis is important part of data mining. It is an unsupervised learning process and it doesn't need prior knowledge about data set. Clustering algorithm is the process that put objects into different clusters according to the attribute of objects and makes objects in one cluster have higher similarity and objects in different clusters have slower similarity. Clustering analysis has been used in many field of life.K-Means cluster is classic partitioning Clustering. It is widely used because it can be implemented easily and has high efficiency. But the algorithm has some problems. One problem is that the count of clusters must be decided prior. The other problem is that it dependent initial cluster centre points and it can reach local minimal result easily. Although KHM algorithm resolves the problem that dependent on initial cluster centre point it still has the problem that it need to decide the count of cluster and will reach local minimal value easily. To resolve these problems we present an algorithm called ACAKHM in this paper. We add ant colony algorithm to the clustering algorithm. Ant colony algorithm has many features. It doesn't need the count of clusters and it is the heuristic random search algorithm that can search global optimum result. It is robust and can add to other algorithm easily.The new algorithm takes full advantage of ant colony algorithm and K-harmonic means clustering algorithm. First, the algorithm initially clusters the data set by colony algorithm to obtain the count of clusters and initial clustering result. Then the cluster centre points derived from ant colony algorithm are used as cluster centre points of K-means algorithm and choose the better initial value to gain the purpose of obtaining the optimum result. The result of experiment indicate that the new algorithm efficiently resolves the problems of KHM algorithm that the count of clusters need decide prior and it well reach local optimum result.