A Bayesian network based framework for real-time crash prediction on the basic freeway segments of urban expressways

► Presents a new framework for real-time crash prediction model building. ► Variables: congestion index in upstream and downstream. ► Variables: speed and occupancy difference between upstream and downstream. ► Method: random multinomial logit (variable selection), Bayesian network (model). ► Performance: 66% crash detection rate with less than 20% false alarm.