An Improved Clustering Approach for Identifying Significant Locations from Spatio-temporal Data

The rapid development and adoption of Internet-of-Things (IoT) and sensors such as Global Positioning System (GPS) in our daily life allows gathering of wealth of information. This tremendous amount of sensor data can be clustered to infer information about the whereabouts of users. Researchers have proposed various algorithms for analyzing spatio-temporal data and clustering such datasets. However, a big challenge is to discover clusters with large density variation and to solve this problem most of the existing clustering algorithms manually set the input parameters. In the Density-based spatial clustering (DBSCAN) clustering algorithm, the epsilon and minimum point parameters have to be set manually for further computation. In this paper, we propose an improved DBSCAN algorithm to adaptively set minimum point parameter for dynamic length of data, based on merge sort and silhouette analysis. Validation of the proposed algorithm has been done on GeoLife GPS dataset. Experimental results show that the proposed algorithm is competitive with state-of-the-art methods for identifying users’ significant locations and whereabouts with the density variation. The paper also aims to highlight the issues pertaining to the invasion of user privacy by the potential application of the clustering algorithm on the aforementioned data.