A BOTTOM-UP HIERARCHICAL CLUSTERING ALGORITHM WITH INTERSECTION POINTS

A pattern classification problem that does not have labelled data points requires a method to assort similar points into separated clusters before the training and testing can be performed. Clustering algorithms place most similar data points into one cluster with highest intra-cluster and lowest inter-cluster similarities. Purpose of this paper is to suggest a bottom-up hierarchical clustering algorithm which is based on intersection points and provides clusters with higher accuracy and validity compared to some well-known hierarchical and partitioning clustering algorithms. This algorithm starts with pairing two most similar data points, afterwards detects intersection points between pairs and connects them like a chain in a hierarchical form to make clusters. To show the advantages of paring and intersection points in clustering, several experiments are done with benchmark datasets. Besides our proposed algorithm, seven existing clustering algorithms are also used. Purity as an external criterion is used to evaluate the performance of clustering algorithms. Compactness of each cluster derived by clustering algorithms is also calculated to evaluate the validity of clustering algorithms. Eventually, the results of experiments show that in most cases the error rate of our proposed algorithm is lower than other clustering algorithms that are used in this study.