Extracting Agent-Based Models of Human Transportation Patterns Rahmatollah Beheshti Department of EECS University of Central Florida Orlando, Florida 32816 Email: beheshti@knights.ucf.edu Gita Sukthankar Department of EECS University of Central Florida Orlando, Florida 32816 Email: gitars@eecs.ucf.edu Abstract—Due to their cheap development costs and ease of deployment, surveys and questionnaires are useful tools for gathering information about the activity patterns of a large group and can serve as a valuable supplement to tracking studies done with mobile devices. However in raw form, general survey data is not necessarily useful for answering predictive questions about the behavior of a large social system. In this paper, we describe a method for generating agent activity profiles from survey data for an agent-based model (ABM) of transportation patterns of 47,000 students on a university campus. We compare the performance of our agent-based model against a Markov Chain Monte Carlo (MCMC) simulation based directly on the distributions fitted from the survey data. A comparison of our simulation results against an independently collected dataset reveals that our ABM can be used to accurately forecast parking behavior over the semester and is significantly more accurate than the MCMC estimator. I. I NTRODUCTION Agent-based simulations have been used successfully for modeling human social systems in diverse fields including economics, sociology, anthropology, and archaeology [1]. A perennial question that arises in the development of an agent- based simulation is how to initialize the models to create a realistic population of agents. In simple models with few parameters, it is feasible to perform a sensitivity analysis to explore the effects of the parameters on the performance of the simulation. However in more complicated agent decision- making models, creating a realistic population of agents can be challenging due to the larger range of parameters governing the behavior of the simulated entities. Surveys and questionnaires can be used to collect an accu- rate static snapshot of the behavior of large social systems but lack the predictive power of simulations. It is more difficult to explore “what-if” questions with a survey since posing questions to participants about hypothetical scenarios can be problematic due to human cognitive biases such as anchoring or risk-aversion. In this paper, we show how both methodolo- gies, surveying and agent-based simulation, can be combined to model human social systems with higher verisimilitude and to explore the ramifications of different behavior patterns and trends. This paper specifically addresses the problem of creating individual agent profiles for an activity-based microsimula- tion model of transportation, dining, parking, and building occupation preferences on a large university campus. One problem with agent-based models is that linking the models and simulation processes with the observed data is challenging. The main contribution of our research is to demonstrate a procedure for systematically linking the observed survey data of people’s transportation preferences with an executable agent model. In contrast, stochastic simulation approaches such as Markov Chain Monte Carlo (MCMC), have been used to forecast the outcome of temporal processes and are simple to create and initialize from observed data [2]. However, in our results, we show that our method is substantially more accurate at forecasting future effects than an MCMC estimator initial- ized from the same survey data, even at answering relatively simple questions. An additional benefit is that manipulating the operation of an agent-based model can empower researchers with better intuitions about the reasons behind emerging group phenomena rather than merely observing the unfolding of a stochastic process [3]. Urban simulation is a particularly fertile area for agent- based simulation research since it requires modeling a large number of interdependent agents making sequential decisions within a small region. Benenson et al. [4] present two moti- vations for defining urban agents, as a distinct group within the general class of autonomous agents: 1) urban agents often have a high degree of mobility resulting in rapidly changing spatial relationships. 2) to succeed, urban agents require a strong capability to perceive and adapt to the evolving urban environment shaped by neighboring agents. In a general urban model, there can be many classes of agents—developer agents constructing new buildings, car agents moving in traffic, business agents providing services to customer agents, and land-use agents who own and manage parcels and lots [4]. In our model, we focus on modeling transient activity patterns such as transportation habits, dining preferences, and building occupation times. The goal is to predict the large-scale aggregate activity patterns of thousands of students over the duration of the semester, in contrast to work that has been done on learning individual transportation modality and route preferences using cell phone and GPS data from hundreds of individuals (e.g., the MIT Reality Mining