Implementation of Multilayer Perceptron and Radial Basis Function Neural Networks for Estimating Roundabout-Entry Traffic Flow

Abstract

Roundabouts capacity is a critical aspect when assessing the feasibility of constructing them. This research examines 50 entry lanes of 15 roundabouts in Greece, both single-lane and multilane. Traffic flows, geometric parameters, and gap acceptance parameters were measured and calculated based on field observations. A quadcopter unmanned aerial vehicle (UAV), RTK GNSS receiver, and video camera attached to a tripod were used to perform the field surveys. Photogrammetry techniques were used to extract the data required for the analysis. The development and evaluation of capacity prediction models involve the implementation of both multilayer perceptron (MLP) and radial basis function (RBF) neural networks. Based on the findings, the current models of Greek and international standards overestimate roundabout capacity. The developed MLP model predicts the existing entry capacity more accurately compared to the RBF model. The developed model can be generalized and the evaluation metrics (R2=0.87 and RMSE=138) indicate that its predictive ability is quite high.