The Virtual City
By Hamdi kavak
Have you ever thought how far you travel on a normal day or how many people you encounter? Research shows that we travel 40 miles on an average day (urban US) and encounter 17 people . These numbers may not mean a lot to an individual… However, when you multiply that by the number of people living in an urban area, you begin to realize that mobility is a major part of urban life. Mobility is so critical part of our life in that it brings unique challenges that we don’t see in rural areas. Challenges like longer lines at the grocery store, sitting in traffic, more gas consumption, air pollution, and perhaps a horrible flu season.
The Virtual City is a step in tackling similar mobility-related challenges and helping provide insights for decision-makers in evaluating their urban plans and make smart policy decisions. The Virtual City is a simulation platform that uses machine learning techniques and real-world mobility data (footprints) to understand the movement of everyday people. What makes this research special is that it extracts movement behavior directly from mobility data, unlike other simulation systems that are based on naive rules.
“The Virtual City is a simulation platform that uses machine learning techniques and real-world mobility data (footprints) to understand the movement of everyday people.
Simply put, geo-located data is collected from publicly accessible sources like call detail records, publicly shared pictures, or GPS-loggers. I used Twitter messages. The Twitter dataset contains around 716 million location footprints of over 6 million people. If you enabled location share in your tweets in the last three years, it is likely that my dataset has your tweets. I want to note that, necessary caution is taken to keep people’s data anonymous.
I developed a simulation of a hypothetical flu season and traced where simulated people got infected the most. The red spots in the middle panel show areas in Chicago that are most likely to be affected by the flu epidemic. As a remedy, health officials could make vaccines more accessible in those areas rather than targeting everywhere in the same way. This is just one use-case; the Virtual City can be used in different contexts. For instance, we can determine the best place to build a new shopping mall that would minimally impact the traffic flow. We can explore the housing in an area based on a new highway construction.
The concept of the Virtual City is based on a data-driven agent-based simulation approach. The user provides mobility and geographic data to be used in semi-automated generation of a simulation and ask policy questions. Simulation results provide insights into what could potentially happen in the real world
The Virtual City emphasizes the importance of human mobility and showcases its impact on our lives personally and as a society. The ultimate goal is to help officials make smart policy decisions and improve the quality of urban life for everyone. This study aims to get us one step closer towards making fully empirically-grounded simulations policy making.
What is Machine Learning?
Machine learning is a research area that focuses on creating computational models and algorithms to extract patterns in data. The Virtual City uses machine learning to capture how individuals move from place to place using people’s own history of location footprints. Identified patterns will provide simulation input and basis for semi-autonomous validation of models.
Machine learning in the Virtual City: example location footprints of an individual is converted to an intermediate format and is used as input to the machine learning model; time-based location prediction is obtained as a result.
The Virtual City was presented by Hamdi Kavak at the 3MT Thesis Challenge Finals at Old Dominion University on October 24, 2017.
Hamdi Kavak (@hmdkvk) studies human behavior using agent-based simulations, statistical models, and machine learning techniques. He is a Ph.D candidate in the department of Modeling, Simulation and Visualization Engineering at Old Dominion University and a graduate research assistant at the Virginia, Modeling, Analysis and Simulation Center (VMASC).