Study Considered Anomaly Detection in Complex Human Behavior by Modeling GPS Data
Human mobility and trajectory modeling have been studied extensively, showcasing varying abilities to manage complex inputs and balance tradeoffs between performance and efficiency. In a new study, researchers developed a model that used GPS data to detect anomalies in human mobility. The model significantly outperformed existing approaches.
The study, by researchers at Carnegie Mellon University and Novateur Research Solutions, is published in ACM Digital Library.
“Modeling human mobility and detecting anomalies are important tasks for an array of applications, from security and surveillance to public health,” notes Leman Akoglu, associate professor of information systems at Carnegie Mellon’s Heinz College, who coauthored the study.
Akoglu and her coauthors considered anomaly detection in complex human behavior by modeling raw GPS data as a sequence of stay-point events, each characterized by spatio-temporal features, along with commutes between the stay-points.
They developed a model that explicitly quantified uncertainty for event sequence modeling, analyzed what these uncertainty measures reveal about human behavior, and incorporated them into uncertainty-aware anomaly detection. The model, called USTAD (for uncertainty-aware spatio-temporal anomaly detection), integrated uncertainty into its mechanism to score anomalies, providing a more nuanced and accurate approach to detecting inconsistencies in human behavior.
Based on experiments on three industry-scale datasets, each containing tens of thousands of agents and millions of events simulated in real-world cities, USTAD consistently achieved superior performance across both event-level and agent-level anomaly detection. On event-level tasks, USTAD improved detection by 2.2% to 165%.
“While previous efforts in detecting human mobility and anomalies have made significant strides, none of them have addressed the challenges in this work as we did,” explains Haomin Wen, a postdoctoral student at Carnegie Mellon’s Heinz College, who coauthored the study.
“Our problem formulation allowed us to leverage modern sequence models for unsupervised training and anomaly detection,” notes Shurui Cao, a PhD student at Carnegie Mellon’s Heinz College, who coauthored the study.
The study was supported by the Intelligence Advanced Research Projects Activity of the U.S. Department of Interior/Interior Business Center.
###
Summarized from an article in ACM Digital Library, "Uncertainty-Aware Spatio-Temporal Human Mobility Modeling and Anomaly Detection," by Wen, H (Carnegie Mellon University), Cao, S (Carnegie Mellon University), Rasheed, Z (Novateur Research Solutions), Shafique, KH (Novateur Research Solutions), and Akoglu, L (Carnegie Mellon University). Copyright 2025 The Authors. All rights reserved
About Carnegie Mellon University's Heinz College of Information Systems and Public Policy
The Heinz College of Information Systems and Public Policy is home to two internationally recognized schools: the School of Information Systems and Management and the School of Public Policy and Management. Heinz College leads at the intersection of people, policy, and technology, with expertise in analytics, artificial intelligence, arts & entertainment, cybersecurity, health care, and public policy. The college offers top-ranked undergraduate, graduate, and executive education certificates in these areas. Our programs are ranked #1 in Information Systems, #1 in Information and Technology Management, #8 in Public Policy Analysis, and #1 in Cybersecurity by U.S. News & World Report. For more information, visit www.heinz.cmu.edu.