Density and neighbor Adaptive Information Theoretic Clustering

This work presents a novel clustering algorithm, named Adaptive Information Theoretic Clustering (AITC). Specific adaptations concerned in AITC are densities and neighbors. Based on the utilization of the within/between information potential, the proposed algorithm is easily computable and carries an intuitive interpretation. We also propose two ways in implementations, the direct and indirect ones, which can not only provide a lower degree of complexity compared with conventional hierarchical clusterings, but also facilitate the adjustment of parameters. Experiments to evaluate the performance of AITC are presented on both synthetic and real datasets with different types of distributions. Better results are gained by the proposed algorithm in comparison with other widely used clustering algorithms.