MULTIOBJECTIVE LEMURS OPTIMIZER ALGORITHM FOR EFFICIENT DBSCAN

Abstract

Clustering algorithms are frequently employed in the fields of data processing as unsupervised learning algorithms. Density-based spatial clustering of applications with noise (DBSCAN), a typical method based on density clustering, can build clusters by finding densely populated regions divided by sparsely populated regions based on cluster density. Moreover, it can effectively cluster unusual data and can function well for any arbitrary-shaped clusters. However, the DBSCAN algorithm has an inherent flaw that cannot be avoided. Because the clustering performance is highly sensitive to the DBSCAN parameters settings i.e. Eps and Minpts, there is no theory to guide the setting of these parameters. To optimize the settings of these parameters, this study proposed a hybrid algorithm that combines the DBSCAN method with the Multiobjective Lemurs Optimizer (MOLO). This method approaches the matter of clustering as a multiobjective optimization challenge to minimize particular cluster validity indices, expressed as objective functions characterizing the quality of the clustering solutions. This made it feasible to determine the correct values for the DBSCAN parameters. The outcomes indicate that the suggested MOLO-DBSCAN is still effective in achieving the most accurate settings for these parameters.

Keywords: DBSCAN, Multiobjective Optimization, Unsupervised Learning, Cluster Validity Indices