Logical analysis of built-in DBSCAN Functions in Popular Data Science Programming Languages

Abstract

DBSCAN algorithm is a location-based clustering approach; it is used to find relationships and patterns in geographical data.  Because of its widespread application, several data science-based programming languages include the DBSCAN method as a built-in function. Researchers and data scientists have been clustering and analyzing their study data using the built-in DBSCAN functions. All implementations of the DBSCAN functions require user input for radius distance (i.e., eps) and a minimum number of samples for a cluster (i.e., min_sample). As a result, the result of all built-in DBSCAN functions is believed to be the same. However, the DBSCAN Python built-in function yields different results than the other programming languages those are analyzed in this study. We propose a scientific way to assess the results of DBSCAN built-in function, as well as output inconsistencies. This study reveals various differences and advises caution when working with built-in functionality.