MDST-DBSCAN: A Density-Based Clustering Method for Multidimensional Spatiotemporal Data

Changlock Choi,Seong-Yun Hong

doi:10.3390/ijgi10060391

Changlock Choi, Seong-Yun Hong

Open Access

https://doi.org/10.3390/ijgi10060391

Copy DOI

Abstract

The increasing use of mobile devices and the growing popularity of location-based ser-vices have generated massive spatiotemporal data over the last several years. While it provides new opportunities to enhance our understanding of various urban dynamics, it poses challenges at the same time due to the complex structure and large-volume characteristic of the spatiotemporal data. To facilitate the process and analysis of such spatiotemporal data, various data mining and clustering methods have been proposed, but there still needs to develop a more flexible and computationally efficient method. The purpose of this paper is to present a clustering method that can work with large-scale, multidimensional spatiotemporal data in a reliable and efficient manner. The proposed method, called MDST-DBSCAN, is applied to idealized patterns and a real data set, and the results from both examples demonstrate that it can identify clusters accurately within a reasonable amount of time. MDST-DBSCAN performs well on both spatial and spatiotemporal data, and it can be particularly useful for exploring massive spatiotemporal data, such as detailed real estate transactions data in Seoul, Korea.

Highlights

The increasing use of mobile devices and the growing popularity of location-based services have generated massive amounts of spatiotemporal data over the last several years. While this phenomenon provides new opportunities to enhance our understanding of various urban dynamics [1,2,3], it poses challenges due to the complex structure and large-volume of the spatiotemporal data
Cluster analysis is useful for this purpose, and various clustering methods have been developed since the 1950s [7,8,9,10,11,12]
We aim to develop an alternative clustering approach that can work for multidimensional spatiotemporal data efficiently and in an explainable fashion

Summary

Introduction

The increasing use of mobile devices and the growing popularity of location-based services have generated massive amounts of spatiotemporal data over the last several years While this phenomenon provides new opportunities to enhance our understanding of various urban dynamics [1,2,3], it poses challenges due to the complex structure and large-volume of the spatiotemporal data. These data often include variables other than time and spatial coordinates to describe the status of an object or a phenomenon [4], and the processing and analysis of such spatiotemporal data require considerable effort and time. Some work has been done on hybrid approaches that combine state-of-the-art deep learning techniques with traditional clustering methods [17]

Objectives

Methods

Conclusion