AF-DBSCAN: An unsupervised Automatic Fuzzy Clustering method based on DBSCAN approach

Automatic clustering problems play an important role to ameliorate the goodness of the data set's partitioning. Actually, the requirement to detect the suitable clustering solution without need for user-given parameters still remain challenging in unsupervised learning. This paper proposes an efficient and effective clustering method, named AF-DBSCAN (Automatic Fuzzy DBSCAN) based on the fuzzy clustering method FN-DBSCAN (Fuzzy Neighborhood Density-Based Spatial Clustering of Applications with Noise). The main idea of the proposed method is to cover the limitations of FN-DBSCAN by exploiting the benefits of k-neighbors plot, in purpose to determine the input parameter values. In fact, AF-DBSCAN avoids the manual intervention of non-experimental users in estimating the input parameters, the minimal threshold of neighborhood membership degree ϵ1 and the minimal neighborhood set cardinality ϵ2, which are hard to guess, and so permits to determine them more reasonably. In such way, the whole clustering process can be fully automated. Simulation experiments, carried out on a real medical data set, highlighted the AF-DBSCAN's effectiveness even for high-dimensions data sets, and showed that the proposed method outperformed the classical method since it provides a better clustering accuracy.