Calibration and Validation of Probabilistic Discretionary Lane-Change Models

Lane changes (LCs) are important in traffic flow operations. They cause differences in flow over lanes and in some cases determine the start of congestion. Whereas calibration and validation are commonly used with car-following models, this is not common practice with LC models. Even then, it is not clear what calibration and validation entails for probabilistic LC models. Therefore, this paper reviews methodologies to calibrate and validate probabilistic LC models, both microscopically and macroscopically. A likelihood is often used in calibration but does not intuitively show the quality of the model. An example shows that it is possible to have the model calibrated and validated with accurate parameters all having the same error in the validation as in the calibration, but the quality of the model is still bad. Using a likelihood ensures the stochastic effects are well captured, but the conclusion is that for validation purposes, one can better use a measure that has physical interpretation and gives a value indicating the quality of the model for the purpose for which it needs to be used.