Modeling Operating Speeds on Residential Streets with a 30 km/h Speed Limit: Regression versus Neural Networks Approach

Abstract

This paper models operating speeds on residential streets with a 30 km/h speed limit by using: (i) regression methods including Single Equation Regression (SER) and Simultaneous Equation Approach (SEA), and (ii) Neural Networks (NN) modeling technique. Free-flow profile-speed data were recorded on 99 street sections with varying characteristics which were then used to develop and validate speed models for estimating maximum speeds obtaining within a section and speeds at the entrance to the next non-signalized intersection. The results suggest that the models developed by SEA performed better than those by the conventional regression (i.e., SER). Compared to regression models, NN models showed better performance especially regarding model fitness although the resultant models are quite complicated. Based on the developed models, various street features were found as determinants of driving speeds that provided helpful information for addressing speeding issues on neighborhood streets.