Identifying best predictors for car following behaviour from empirical data

A car-following model has been developed which represents an open-loop input-output mapping expressed either as a deterministic car-following model, or a fuzzy logic model with arbitrary precision. The model was used to relate the driver's action (output) with the behaviour predictors (inputs) he or she can perceive. Time-series data of motorway car following process were used to train the proposed model under different inputs and time delay assumptions. It was found that relative speed related inputs, combinations of relative speed related and headway related inputs all gives satisfactory performance. The best predictors for car-following behaviour were identified. The behavioural implication of identified predictors was further explored. It was found that driver used speed-matching and headway-matching as the main strategy in performing car-following task However, drivers seems choose to use different predictors which is easy to perceive in different traffic situation. The methodology used in this research is a novel, which made it possible to observe dynamic behaviour as a continuous process instead of a series of discrete events. The finding is believed to have practical implication in traffic modelling and driver behaviour research.