Optimizing the Ant Clustering Model Based on K-Means Algorithm

Ant clustering is one of effective clustering methods. Compares to other clustering methods, ant clustering algorithm has one outstanding advantage and one disadvantage. The advantage is that the total numbers of cluster is generated automatically ,and the disadvantage is that its cluster result is random and its result is influenced by the input data and the parameters, which leads low quality of its cluster result. In this paper, we propose an improved ant clustering algorithm based on K-means, which optimizes the rules of ant clustering algorithm. In our system, we also decide the proper values of parameters Pdel and Iter by training the training datasets before we cluster. Experimental results demonstrate that the proposed method has a good performance.

1. Introduction

Clustering is a process of grouping a set of data objects into clusters based on the information found in the data objects^[1] After completing the clustering process, data objects in a cluster are similar to each other and are different from data objects in other clusters. Because the grouping phenomenon for data objects can be captured through the clustering process, clustering plays an important role in various data analysis fields including statistics (McLachlan and Krishnan, 1997), pattern recognition (Webb, 2002), machine learning (Alpaydin, 2004), data mining (Tan et al., 2005),information retrieval (Wu et al., 2003), and bioinformatics (He et al., 2006). Among various clustering algorithms, K-means(Forgy, 1965; McQueen, 1967) is one popular and widespread partitioning algorithm because of its superior feasibility and efficiency in dealing with a large amount of data (Hand and Krzanowski, 2005). Ant clustering algorithm is another useful clustering algorithm because it is able to find utomatically a good partition over artificial and real data sets. Furthmore, it does not need the number of expected clusters to converge(Nicolas Labroche, 2002). Ant clustering can treat small to big sets of data with a great success but also demonstrate, that Ant clustering does not manage to find a good partition when an important number of clusters is expected. This may be due to the fact that there is only one rule that can create a new nest^[2].