Assessing Weather,Environment,and Loop Data for Real-Time Freeway Incident Prediction

Abstract

Weather, environment, and loop data conditions are promising indicators for real-time freeway incident prediction. The ability to predict the likelihood of selected incident types by using weather and environment data was examined. Loop detector data were analyzed for conditions useful for in-lane incident prediction. Nonnested and nested multinomial logit models were estimated with data from selected freeways in Austin, Texas. The estimation results revealed that factors such as visibility, time of day, and lighting condition are significant determinants of incident type, whereas 5-min average occupancy and coefficient of variation in speed are strong predictors of in-lane freeway accidents.