Accelerated DBSCAN via parallel, density-aware multi-objective genetic optimization

Clustering is a fundamental task in data mining, where the quality of results often hinges on effective parameter selection. DBSCAN is widely used for discovering clusters of arbitrary shapes but is highly sensitive to its input parameters \textit{Eps} and \textit{MinPts}. This paper proposes an enhanced version of the Multi-Objective Genetic Algorithm for DBSCAN, termed \textbf{Enhanced MOGA-DBSCAN}, which introduces a modified objective function based on a density-aware Outlier Index and accelerates the optimization process through parallel computation. We evaluate the proposed method using two benchmark datasets and compare it against the original MOGA-DBSCAN as well as two adaptive variants: AMD-DBSCAN and WOA-DBSCAN. Results show that Enhanced MOGA-DBSCAN consistently achieves superior clustering performance, as measured by Rand Index and Normalized Mutual Information (NMI), while also reducing runtime relative to the original MOGA-DBSCAN. These findings highlight the effectiveness of our enhancements in improving both clustering quality and computational efficiency.

Keywords:

Machine learning , Unsupervised learning , Clustering , DBSCAN , Genetic Algorithm