Behavior Modeling and Analytics for Urban Computing: A Synthetic Information-based Approach
| dc.contributor.author | Parikh, Nidhi Kiranbhai | en |
| dc.contributor.committeechair | Marathe, Madhav Vishnu | en |
| dc.contributor.committeechair | Swarup, Samarth | en |
| dc.contributor.committeemember | Vullikanti, Anil Kumar S. | en |
| dc.contributor.committeemember | Sukthankar, Gita Reese | en |
| dc.contributor.committeemember | Ramakrishnan, Naren | en |
| dc.contributor.department | Computer Science | en |
| dc.date.accessioned | 2018-09-07T06:00:15Z | en |
| dc.date.available | 2018-09-07T06:00:15Z | en |
| dc.date.issued | 2017-03-15 | en |
| dc.description.abstract | The rapid increase in urbanization poses challenges in diverse areas such as energy, transportation, pandemic planning, and disaster response. Planning for urbanization is a big challenge because cities are complex systems consisting of human populations, infrastructures, and interactions and interdependence among them. This dissertation focuses on a synthetic information-based approach for modeling human activities and behaviors for two urban science applications, epidemiology and disaster planning, and with associated analytics. Synthetic information is a data-driven approach to create a detailed, high fidelity representation of human populations, infrastructural systems and their behavioral and interaction aspects. It is used in developing large-scale simulations to model what-if scenarios and for policy making. Big cities have a large number of visitors visiting them every day. They often visit crowded areas in the city and come into contact with each other and the area residents. However, most epidemiological studies have ignored their role in spreading epidemics. We extend the synthetic population model of the Washington DC metro area to include transient populations, consisting of tourists and business travelers, along with their demographics and activities, by combining data from multiple sources. We evaluate the effect of including this population in epidemic forecasts, and the potential benefits of multiple interventions that target transients. In the next study, we model human behavior in the aftermath of the detonation of an improvised nuclear device in Washington DC. Previous studies of this scenario have mostly focused on modeling physical impact and simple behaviors like sheltering and evacuation. However, these models have focused on optimal behavior, not naturalistic behavior. In other words, prior work is focused on whether it is better to shelter-in-place or evacuate, but has not been informed by the literature on what people actually do in the aftermath of disasters. Natural human behaviors in disasters, such as looking for family members or seeking healthcare, are supported by infrastructures such as cell-phone communication and transportation systems. We model a range of behaviors such as looking for family members, evacuation, sheltering, healthcare-seeking, worry, and search and rescue and their interactions with infrastructural systems. Large-scale and complex agent-based simulations generate a large amount of data in each run of the simulation, making it hard to make sense of results. This leads us to formulate two new problems in simulation analytics. First, we develop algorithms to summarize simulation results by extracting causally-relevant state sequences - state sequences that have a measurable effect on the outcome of interest. Second, in order to develop effective interventions, it is important to understand which behaviors lead to positive and negative outcomes. It may happen that the same behavior may lead to different outcomes, depending upon the context. Hence, we develop an algorithm for contextual behavior ranking. In addition to the context mentioned in the query, our algorithm also identifies any additional context that may affect the behavioral ranking. | en |
| dc.description.abstractgeneral | The rapid increase in urbanization poses challenges in diverse areas such as energy, transportation, pandemic planning, and disaster response. Human activities and behaviors are at the center of such socially-coupled systems. This dissertation focuses on modeling human activities and behaviors for two urban science applications, epidemiology and disaster planning, and with associated analytics. Our approach is based on a synthetic information, which is a detailed representation of human populations, infrastructural systems and their behavioral and interaction aspects. It is created by combining data from multiple sources and used in large-scale simulations to study what-if scenarios and for policy recommendation. Big cities have a large number of visitors visiting them every day. They often visit crowded areas in the city and come into contact with each other and the area residents. However, most epidemiological studies have ignored their role in spreading epidemics. We model transient populations, consisting of tourists and business travelers, for Washington DC metro area. Transients are modeled along with their demographics and activities by combining data from multiple sources. We simulate a flu-like disease outbreak both with and without transient populations to evaluate their effects on spreading epidemics. We also evaluate multiple interventions that target transients. In the next study, we model human behavior in the aftermath of the detonation of an improvised nuclear device in Washington DC. Previous studies of this scenario have mostly focused on modeling physical impact and evaluating whether it is better to shelter-in-place or evacuate. But they do not take into account what people actually do in the aftermath of disasters. Natural human behaviors in disasters, such as looking for family members or seeking healthcare, are supported by infrastructures such as cell-phone communication and transportation systems. We model natural human behaviors such as looking for family members, evacuation, sheltering, healthcare-seeking, worry, and search & rescue and their interactions with infrastructural systems. Large-scale and complex agent-based simulations generate a large amount of data, making it hard to make sense of results. This leads us to study two new problems in simulation analytics. First, we develop algorithms to summarize simulation results by extracting causally-relevant descriptions from agent trajectories. These descriptions are informative about agents’ outcomes at the end of the simulation. Second, in order to develop effective interventions, it is important to understand which behaviors lead to positive and negative outcomes. It may happen that the same behavior may lead to different outcomes, depending upon the context. Hence, we develop an algorithm for contextual behavior ranking. | en |
| dc.description.degree | Ph. D. | en |
| dc.format.medium | ETD | en |
| dc.identifier.other | vt_gsexam:9817 | en |
| dc.identifier.uri | http://hdl.handle.net/10919/84967 | en |
| dc.publisher | Virginia Tech | en |
| dc.rights | In Copyright | en |
| dc.rights.uri | http://rightsstatements.org/vocab/InC/1.0/ | en |
| dc.subject | Behavior Modeling | en |
| dc.subject | Simulation Analytics | en |
| dc.subject | Social Simulations | en |
| dc.subject | Synthetic Information | en |
| dc.subject | Transient Population | en |
| dc.subject | Urban Computing | en |
| dc.title | Behavior Modeling and Analytics for Urban Computing: A Synthetic Information-based Approach | en |
| dc.type | Dissertation | en |
| thesis.degree.discipline | Computer Science and Applications | en |
| thesis.degree.grantor | Virginia Polytechnic Institute and State University | en |
| thesis.degree.level | doctoral | en |
| thesis.degree.name | Ph. D. | en |
Files
Original bundle
1 - 1 of 1