A kernel density smoothing method for determining an optimal number of clusters in continuous data

摘要: While data clustering algorithms are becoming increasingly popular across scientific, industrial and social mining applications, model complexity remains a major challenge. Most do not incorporate mechanism for finding an optimal scale parameter that corresponds to appropriate number of clusters. We propose , kernel-density smoothing-based approach clustering. Its main ideas derive from two unsupervised approaches – kernel density estimation (KDE) scale-spacing (SSC). The novel method determines the clusters by first dense regions in before separating them based on data-dependent estimates. is determined different levels smoothing after inherent arbitrary shape has been detected without priori information. demonstrate applicability proposed under both nested non-nested hierarchical methodologies. Simulated real results presented validate performance method, with repeated runs showing high accuracy reliability.