Fast identification of critical roads by neural networks using system optimum assignment information

Identification of critical segments in a road network is a crucial task for transportation system planners as it allows for in depth analysis of the robustness of the city's infrastructure. The current techniques require a considerable amount of computation, which does not scale well with the size of the system. With recent advances in machine learning, especially classification techniques, there are methods, which can prove to be more efficient replacements of current approaches. In this paper we propose a neural network (NN) based approach for classification of critical roads under user equilibrium traffic (UE) assignment. We, furthermore, introduce a novel predictor attribute, which captures the contrast between UE and system optimum (SO) assignment on the network. Our results demonstrate that the neural network can achieve considerable identification precision of critical road segments and that the SO related attributes significantly increase the classification power. We, furthermore, demonstrate that the NN approach outperforms the commonly used approach of linear regression (LR) and another popular classification approach from the field of machine learning, namely support vector machines (SVM).