Method to Select Variables for Estimating the Parameters of Equations That Describe Average Vehicle Travel Speed in Downtown City Areas

A lack of public vehicular traffic data for a city limits our understanding of the traffic occurring in the street networks of that city; however, there are free tools to extract street network graphs from digital maps and to assess the static properties associated with those graphs. This study proposes a two-stage modeling method to describe dynamic traffic data with static street network features. A quadratic polynomial is used to fit the average travel speed (ATS) pattern observed in the city center. Then, the relationship between the polynomial parameters and street network variables is analyzed through multiple linear regression. Descriptive geometric and topological measurements of downtown areas are obtained with the OSMnx tool (from OpenStreetMap), and with these data, independent variables are defined. The speed of vehicles, assessed every 15 min (from 6:00 a.m. to 10:00 p.m.) on the downtown street networks of twelve major cities, is obtained with the distance_matrix service of GoogleMaps, and with these data, the ATS (the dependent variable) is calculated. The ATS (presenting a U-shape) is modeled with a polynomial equation of order two, so there are three parameters for each city; in turn, each parameter is modeled with a multiple linear regression equation with the independent variables. For training purposes, the ATS equation parameters of ten cities are calculated, and the parameters, in turn, are explained with the proposed method. For validation purposes, the parameters of two cities not considered in the training process are calculated with the multiple linear regression equations. The ATS equation parameters of the twelve cities are correctly modeled so that each city’s ATS can be adequately described. It was concluded that the method selects the independent variables that are suitable to explain the ATS equation parameters. In addition, with the Akaike information criterion, the variable selection case presenting the best trade-off between accuracy and complexity is identified.