A framework based on Natural Language Processing and Machine Learning for the classification of the severity of road accidents from reports

Road safety analysis is typically performed by domain experts on the basis of the information contained in accident reports. The main challenges are the difficulty of considering a large number of reports in textual form and the subjectivity of the expert judgments contained in reports. This work develops a framework based on the combination of Natural Language Processing (NLP) and Machine Learning (ML) for the automatic classification of accidents with the final aim of assisting experts in performing road safety analyses. Two different models for the representation of the textual reports (Hierarchical Dirichlet Processes (HDPs) and Doc2vec) and three ML-based classifiers (Artificial Neural Networks (ANNs), Decision Trees (DTs) and Random Forests (RFs)) are compared. The framework is applied to a repository of road accident reports provided by the US National Highway Traffic Safety Administration. The best trade-off between accuracy of the classification and explainability of the obtained results is achieved by combining HDP topic modeling and RF classification.