Fuzzy joint points based clustering algorithms for large data sets

The fuzzy joint points (FJP) method is one of the successful fuzzy approaches to density-based clustering. Besides the basic FJP method, there are other methods based on the FJP approach such as, Noise-Robust FJP (NRFJP), and Fuzzy Neighborhood DBSCAN (FN-DBSCAN). These FJP-based methods suffer from the low speed of the FJP algorithm, thus applications that deal with large databases cannot benefit from them. The Modified FJP (MFJP) method addresses this issue and achieves an improvement in speed, but it is not satisfactory from the point of applicability. In this work, we integrate various methods with FJP to establish an optimal-time algorithm. An even faster algorithm which uses the FJP approach in a somewhat supervised fashion is also proposed. Along with theoretic comparison, experimental results are presented to show the significant speed improvement, which will allow the FJP-based methods to be used on large data sets.