Development of a Neural Network Pedestrian Gap Acceptance Model for Enhanced Operation of Midblock Signalled Pedestrian Crossings

UK transport policy has shifted dramatically in recent years. The new policy direction seeks to promote walking as an alternative to the car for short trips. Midblock signalled pedestrian crossings are a common method of resolving the conflict between pedestrians and vehicles. This paper considers alternative operating strategies for midblock signalled pedestrian crossings that are more responsive to the needs of pedestrians without adding to the costs of business by increasing delay to motorists and freight traffic. Specifically the paper considers the development of a pedestrian gap acceptance model that would be required in the proposed strategies. A succession of artificial neural network models are developed and factors influencing performance of the models both in terms of accuracy and processing requirements are considered in detail. The paper concludes that a feedforward artificial neural network using backpropagation can deliver a gap acceptance model with a high degree of accuracy within the context and constraints under which it has been developed.