An Improved Agglomerative Clustering Method

Abstract

Clustering is a common and useful exploratory task widely used in Data mining. Among the many existing clustering algorithms, the Agglomerative Clustering Method (ACM) introduced by the authors suffers from an obvious drawback: its sensitivity to data ordering. To overcome this issue, we propose in this paper to initialize the ACM by using the KKZ seed algorithm. The proposed approach (called KKZ_ACM) has a lower computational time complexity than the famous kmeans algorithm. We evaluated its performance by applying on various benchmark datasets and compare with ACM, kmeans++ and KKZ_ k-means. Our performance studies have demonstrated that the proposed approach is effective in producing consistent clustering results in term of average Silhouette index. General Terms Data mining, Algorithms