Detection of hierarchical crowd activity structures in geographic point data

Area of Science:

• Geographic Information Science
• Computational Science
• Data Mining

Background:

• The proliferation of GPS sensors generates vast amounts of geolocated event data, crucial for understanding social interactions.
• Existing analytical tools often lack the sophistication to fully leverage the granularity of this data for complex spatial analysis.
• Event data, characterized by point signals in space and time, is used for tasks like anomaly detection and land use extraction.

Purpose of the Study:

• To develop a unified framework for detecting hierarchical crowd activity structures in geographic point data.
• To bridge computational science and geographical analysis by integrating hierarchical scale concepts with clustering techniques.
• To introduce novel methods for synthetic data generation, DBSCAN parameter selection, and algorithm evaluation for spatial data.

Main Methods:

• Generation of synthetic geographic event data mirroring real-world distributions.
• An improved density-based spatial clustering of applications with noise (DBSCAN) algorithm with automatic parameter selection for iterative analysis.
• Development of a framework for evaluating algorithms designed to extract hierarchical structures from spatial data.

Main Results:

• The proposed approach successfully generates representative synthetic geographic event data.
• The enhanced DBSCAN algorithm effectively uncovers hierarchical structures in event databases.
• The evaluation framework provides a robust method for comparing crowd activity detection algorithms.

Conclusions:

• The developed framework is effective for comparing crowd activity detection algorithms.
• The automatic DBSCAN parameter selection offers a novel method for identifying hierarchical patterns in geographic point data.
• This research advances the analysis of large-scale geolocated social interaction data.