A Clustering Algorithm for Varied Density Clusters based on Similarity of Local Density of Objects

摘要

Cluster analysis aims to discover the hidden relationship between objects in the dataset so that similar objects are placed in the same cluster and non-similar ones are placed in different clusters. This research proposes a clustering algorithm for datasets that contain diverse clusters in density; the cluster is a connected graph where the similarity between any two adjacent neighbors is greater than or equal to a threshold. The similarity is based on the local density of objects, where the local density of an object is the sum of the distances between it and its k-nearest neighbors. The strategy starts from any object to collect its similar objects from its neighbors and continues collecting similar objects for the collected neighbors until there is no similar object can be added to the current cluster. The suggested strategy is tried on to two synthetic datasets and many reference datasets used in this field, which are available on http://cs.joensuu.fi/sipu/datasets/. All of them have two dimensions to easily visualize the result. The results uncover the effectiveness of the proposed strategy in determining groups with varied forms, sizes, and densities from the given datasets.