An Improved MOGA-DBSCAN with Voronoi-Derived Epsilon for Robust Clustering

Clustering is a fundamental task in data mining and machine learning, grouping similar objects based on their features. It has applications in image analysis, market segmentation, and bioinformatics. Among clustering methods, DBSCAN (Density-Based Spatial Clustering of Applications with Noise) stands out for finding arbitrary-shaped clusters and handling noise. DBSCAN requires two parameters: ϵ (neighborhood radius) and MinPts (minimum points for a dense region). Accurate parameter determination is crucial for algorithm performance. This paper proposes a novel approach to DBSCAN parameter determination. Instead of treating ϵ as continuous, we discretize it using radii of empty circles from the Voronoi diagram of the points. This simplifies ϵ selection and enhances clustering robustness. Our method leverages Voronoi geometric properties, offering a more intuitive and accurate way to set ϵ. Experimental results show that this discrete approach simplifies parameter tuning while maintaining or improving clustering quality compared to existing methods. This advancement enables more reliable and efficient clustering in practice.

کلیدواژه ها:

Clustering ، DBSCAN ، Genetic Algorithm ، Optimization ، Voronoi Diagram ، Delaunay Triangulation