This paper provides a new approach for driving behavior; driving is not anymore a discrete process but a story divided into multiple episodes. Thiscontinuous story is characterized by a probability of terminating a givenepisode based on the driving experience encountered. To capture the above approach in a quantifiable framework, duration modeling is used (1) Duration models are frequently encountered in the domains of economics, socialsciences, political studies and even, in transport (2) However, due to the complexities involved in specifying the boundary conditions of the problem, duration modeling was never applied in traffic. To overcome this difficulty, extensive analysis and modifications was performed on the NGSIM data (3) detailed trajectory data is transformed into episode duration data. The extracted episode-based data was studied and two final duration modelswere found: A Weibull Duration Model and a Log-Logistic Duration Model. Both models suggest that the probability of changing lanes (hazard functions) increases at the beginning of the episode (impatience or snow-balling effect). The hazard function starts decreasing afterward (habit or inertia effect). Moreover, it was shown that the behavior of the follower and the second leader have a significant role in the decision of the driver when changing lanes. This proves that driving is an anisotropic phenomenon whereanticipation plays an important role.
Abstract