Theory And Practice Of Ant Clustering And Partitioning-based DBSCAN Clustering

Clustering problem is an unsupervised learning problem. It is a procedure that cluster data objects into matching clusters. The data objects in the same clusters are quite similar to each other and dissimilar in the other clusters.Density-based clustering defines cluster as region, the objects of the region are dense. These clusters are separated from one another by low-density regions. Density-based clustering algorithms have significant advantages over partitioning and hierarchical clustering algorithms. It can discover clusters of arbitrary shapes. In addition it is able to effectively identify noise pointsDBSCAN algorithm is one of the density-based clustering algorithms. It has the same characters as density-based clustering. But it is sensitive to the input parameters and dimension of data. In addition, the algorithm needs large volume of memory support and has difficulty with clusters of very different densities. So, partitioning-based DBSCAN algorithm (PDBSCAN) was proposed to solve these problems. But PDBSCAN will also get poor results when the density of data is non-uniform.In this paper, we propose a new algorithm (PACA-DBSCAN) to solve these problems using partitioning-based DBSCAN and ant clustering algorithm. We use modified ant clustering algorithm (ACA) and design a new partitioning algorithm based on'point density'(PD) in data preprocessing phase. The experiment results on five data to indicate the superiority of the PACA-DBSCAN algorithm. The performance of the PACA-DBSCAN algorithm is compared with the DBSCAN, PDBSCAN, ACA and KHM algorithm.