A Hybrid Clustering Algorithm: The FastDBSCAN

Abstract

Clustering is one of the most important tasks in knowledge discovery from data. The goal of clustering is to discover the nature structure of data or detect meaningful groups from data. The density-based clustering, such as DBSCAN, is a fundamental technique with many advantages in applications. However, DBSCAN algorithm has the quadratic time complexity, making it difficulty in real application with large data set. This paper presents a method to decrease the time complexity based on K-Means algorithm. Our algorithm divides the data in k partitions at first step and then uses a Min-Max method to select points for DBSCAN clustering at second step. Experiments show that our method obtains competitive results with the original DBSCAN, while significantly improving the computational time.