Using Generative Adversarial Network in predicting people's travel based on socio-economic characteristics

Over the years, due to increasing costs of new infrastructure and increasing concerns about traffic congestion and air pollution, the dimensions of supply-based transportation planning have broadened and included rout access problems as well as travel demand management. As a result, there are travel demand management strategies such as congestion pricing, telecommuting, e-government, etc. that try to change people's travel patterns and manage travel demand. Travel demand management policies have led to a change in the focus of travel modeling from travel-based forecasting (four-step models) in the long-term time frame and at the aggregate level to travel modeling at the disaggregate level (individual level). For this purpose, in this study, the travel and activity pattern of people is predicted based on their socio-economic characteristics. In this study, data from the city of Washington, which was collected in 2007/2008, was used. This prediction was made using the generative adversarial network (GAN) machine learning model that received people's socio-economic information such as age, gender, income, employment status, workplace, vehicle ownership and household size as input and predicts which of the six activities of home, work, education, shopping, recreation, and other people do every half hour of a day. For validate the models FID, chi square statistic, time pattern diagram, comparison of the total number of activities and R-square have been used. Based on the validations, the adversarial generative network model was able to predict people's activities in a day with high accuracy.