Andrew W. Frederiksen

DBSCAN is a widely used unsupervised machine learning algorithm for clustering and spatial data analysis. However, the accuracy of the algorithm is highly dependent on the selection of its hyperparameters, minimum samples (Smin), the minimum number of points required to form a cluster, and ϵ, the maximum distance between points. In this study, we propose a modification to the DBSCAN algorithm by introducing an event density map replacing Smin and ε. Through this method, we decrease the number of hyperparameters from two to one, N which represents the number of cells in the event density map, simplifying, and speeding up optimization. As a result, the optimization of the algorithm will be improved as the sole factor to consider is the size of each cell. In addition, the utilization of dynamic Smin will provide more effective clustering because it is better suited to regions that have a variable earthquake density. We used the Iranian earthquake catalog for testing the algorithm, and we compared the outcomes to the Mirzaei et al. (1998) model as a standard for evaluation. Because this algorithm allows for density contrasts between clusters, it can be a good indicator when studying the zonation of a new area. The findings were more consistent than those of other methods and may offer additional insight into the seismotectonic of Iran. Other than earthquake studies, this algorithm can be applied in multiple fields of science and engineering for clustering datasets with variable-density clusters.