Enhancing Density-Based Spatial Clustering of Applications with Noise (DBSCAN) with Kernel Density Estimation

Abstract

Density-Based Spatial Clustering of Applications with Noise (DBSCAN) is one of the prominent methods that are efficient at uncovering clusters of various shapes. However, it faces limitations when dealing with datasets containing clusters of varying densities, to address this limitation this study integrates kernel density estimation into the DBSCAN algorithm to enhance its capacity to capture density variations and handle irregularly shaped clusters. Specifically, we employ Kernel Density Estimation (KDE) using Epanechnikov as the kernel function and the grid search method with cross-validation for the bandwidth selection, along with the added density threshold. The simulation study shows that the proposed procedures were able to correctly specify the number of clusters even for varying densities. Moreover, empirical results show that the proposed clustering procedure was able to enhance the DBSCAN algorithm and give meaningful results.