Mining traffic incidents to forecast impact

Using sensor data from fixed highway traffic detectors, as well as data from highway patrol logs and local weather stations, we aim to answer the domain problem: "A traffic incident just occurred. How severe will its impact be?" In this paper we show a practical system for predicting the cost and impact of highway incidents using classification models trained on sensor data and police reports. Our models are built on an understanding of the spatial and temporal patterns of the expected state of traffic at different times of day and locations and past incidents. With high accuracy, our model can predict false reports of incidents that are made to the highway patrol and classify the duration of the incident-induced delays and the magnitude of the incident impact, measured as a function of vehicles delayed, the spatial and temporal extent of the incident. Equipped with our predictions of traffic incident costs and relative impacts, highway operators and first responders will be able to more effectively respond to reports of highway incidents, ultimately improving drivers' welfare and reducing urban congestion.