New insights into road accident analysis through the use of text mining methods

Traffic safety is one of the main goals when designing infrastructure or vehicles. Many entities are interested in accident data in order to identify current problems. The accurate recording and analysis of accident data, though, is a time-consuming task. New data mining methods allow for a more efficient analysis of large amounts of (unstructured) data. In this paper, accident data provided by the police department of the city of Munich is analyzed both by classical methods considering only classified accidents, as well as with the help of text mining methods, taking into account all descriptions of the accidents. The results indicate that text mining methods can give quick results identifying main problems at the given locations. Using statistical and machine learning methods for the analysis shows promising results. The results of the automatic classification of accidents into accident types based on a learned classification model using the textual descriptions is highly accurate. Retrieving structured information from the descriptions, though, requires a more concise writing of the accidents reports. We conclude that in future, data mining methods could be used to reduce the workload for police officers for both the reporting work as well as the analysis of road accidents.