Predicting Road Accidents Based on Current and Historical Spatio-temporal Traffic Flow Data

This paper presents research work towards a novel decision support system that predicts in real time when current traffic flow conditions, measured by induction loop sensors, could cause road accidents. If flow conditions that make an accident more likely can be reliably predicted in real time, it would be possible to use this information to take preventive measures, such as changing variable speed limits before an accident happens. The system uses case-based reasoning, an artificial intelligence methodology, which predicts the outcome of current traffic flow conditions based on historical flow data cases that led to accidents. This study focusses on investigating if case-based reasoning using spatio-temporal flow data is a viable method to differentiate between accidents and non-accidents by evaluating the capability of the retrieval mechanism, the first stage in a case-based reasoning system, to retrieve a traffic flow case from the case base with the same outcome as the target case. Preliminary results from experiments using real-world spatio-temporal traffic flow data and accident data are promising.