University of Twente Student Theses

This website will be unavailable due to maintenance December 1st between 8:00 and 12:00 CET.

A graph-based clustering approach for detecting recurring spatio-temperal groupings in event data

Bhat, Rajit Kumar (2021) A graph-based clustering approach for detecting recurring spatio-temperal groupings in event data.

Link to full-text:
(only accessible for UT students and staff)
Abstract:With a proliferation of location-based devices, a large quantum of data is being generated, the nature of which is spatial, temporal or Spatio-temporal (ST). ST data has benefits over purely spatial or temporal data as it simultaneously helps in understanding persistence patterns and highlight unusual patterns over time. It is observed in mobility studies for humans and animals that movements are not random and serve a certain purpose which is fundamental to the dynamics of the respective ecosystems. Thus, understanding the pattern of their recurring nature can give insights into their groupings in both space and time. Clustering can be helpful in interpreting ST data. However, clustering algorithms that use Euclidean or planar distance do not account for constraints offered by space and overestimate clustering. Moreover, existing exploratory methods that have a linear view of time, may not recognise the recurring nature of ST events. Also, the spatial boundary and temporal boundary need to be combined in a meaningful manner. The current work develops an unsupervised exploratory method for detecting recurring Spatio-temporal groupings in ST event data, using a graph-based strategy to agglomerative hierarchical clustering framework. The method is applied to “check-in” data for a location-based social network (LBSN) service to identify clusters or hotspots of users’ activity in a city. Hierarchical clustering does not require pre-selection of clusters, enabling automatic discovery of users based on the Spatio-temporal similarity in check-ins. To incorporate the notion of recurrence, a cyclic view of time has been adopted into the temporal distance. The use of a distance metric for clustering where the spatial and temporal components are merged in a weighted linear combination with appropriate scales allows to conceive the spatial and temporal dimensions as required.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page