Calibration of Safety Prediction Models for Planning Transportation Networks

Abstract

The current planning practice addresses safety implicitly as a by-product of adding capacity and operational efficiency to the transportation system. Safety-conscious planning is a new proactive approach to the prevention of crashes based on establishing inherently safe transportation networks through integrating consideration of safety into the transportation planning process. One of the major concerns in predicting crashes in transportation networks is the applicability and accuracy of crash prediction models. The paper presents two alternative formulations of the calibration problem consistent with the maximum likelihood approach. The proposed formulations can be viewed as a generalized version of the existing calibration procedure proposed in the past for individual crash prediction models. The formulations are useful for road networks and for any transportation mode, provided that the needed prediction models are available. The proposed calibration applied to individual elements of the test network yielded crash estimates that exhibited a considerable bias accumulation at the system level. The calibration task was redefined to focus on the prediction of the cumulative number of crashes in the user-defined subnetworks. The second method gave more acceptable results. The paper demonstrates the feasibility of the proposed approach, which may be helpful in developing a new class of tools for safety-conscious planning.