A Genetic Algorithm-based Method for Improving Quality of Travel Time Prediction Intervals Abbas Khosravi a , Ehsan Mazloumi b , Saeid Nahavandi, Doug Creighton a , J. W. C. Van Lint c a Centre for Intelligent Systems Research (CISR),Deakin University, Geelong, VIC, 3217, Australia b Institute of Transport Studies, Department of Civil Engineering, Monash University, Melbourne, Australia c Department of Transport and Planning, Faculty of Civil Engineering and Geosciences, Delft University of Technology, 2600 Delft, The Netherlands Abstract The transportation literature is rich in the application of neural networks for travel time predic- tion. The uncertainty prevailing in operation of transportation systems, however, highly degrades prediction performance of neural networks. Prediction intervals for neural network outcomes can properly represent the uncertainty associated with the predictions. This paper studies an applica- tion of the delta technique for the construction of prediction intervals for bus and freeway travel times. The quality of these intervals strongly depends on the neural network structure and a train- ing hyperparameter. A genetic algorithm–based method is developed that automates the neural network model selection and adjustment of the hyperparameter. Model selection and parame- ter adjustment is carried out through minimization of a prediction interval-based cost function, which depends on the width and coverage probability of constructed prediction intervals. Ex- periments conducted using the bus and freeway travel time datasets demonstrate the suitability of the proposed method for improving the quality of constructed prediction intervals in terms of their length and coverage probability. Keywords: Travel time, prediction interval, neural network, genetic algorithm. 1. Introduction 1.1. Travel Time Prediction Problem Access to accurate travel time information is widely acknowledged to have the potential to increase the reliability in road networks [1] and to alleviate congestion and its negative environ- mental and societal side effects. From the travelers’ perspective, information on future travel times can reduce the uncertainty in decision making in regard to departure time, route and mode choice [2] [3] [4], which in turn can lessen travelers’ stress and anxiety [5] [6]. From the oper- ators’ point of view, information on future travel times can be used to present the current traffic state in a network and to fully identify the problematic locations/routes [7]. Travel time prediction is a complex problem because travel times result from nonlinear in- teractions of heterogeneous groups of driver-vehicle combinations, each characterized by their own specific technical and behavioral properties, such as vehicle dimensions, acceleration char- acteristics, and driving styles (aggressive or conservative) [8]. Travel times are also influenced by exogenous factors that are often completely beyond the analyst’s capacity to predict, such as weather and traffic incidents. There exists a wide range of methodologies adopted to predict Preprint submitted to Transportation Research Part C April 13, 2011