FREEWAY INCIDENT-DETECTION ALGORITHMS BASED ON DECISION TREES WITH STATES

Incident-detection algorithms are a part of an overall freeway-traffic management system. These algorithms provide indications of the probable presence of freeway incidents by processing electronic surveillance data. In this paper, a class of algorithms that are designed to discriminate patterns in the data peculiar to incidents are described. The generic structure of these algorithms is the decision tree with states, the states corresponding to distinct traffic conditions. Ways to calibrate algorithm thresholds are described and applied to the algorithms. Performance evaluations based on traffic data from the Los Angeles system are presented. /Authors/