Formal Concept Analysis for Traffic Accident Summaries and Construction of Traffic Accident Prediction Model

The purpose of this study is to support police activities targeting elderly pedestrians and other vulnerable road users, so we analyze past traffic accident data using Formal Concept Analysis (FCA) and prediction using machine learning models to prevent future traffic accidents. Traffic accident data contains an accident summary that summarizes the circumstances leading up to the oc-currence of each accident in 30 to 200 words. Since FCA requires a table organized by binary information, words are extracted from the accident summaries using Morphological Analysis (MA) and the TF-IDF method. In addition, to cope with word orthography and synonymy, synonym unification using Word2Vec, etc. is performed after MA. Then, the data organized into two value information is applied to FCA, hypotheses are formulated based on the rules obtained, and an attempt is made to elucidate the causes of traffic accidents through statistical analysis. In addition, the accident summaries are vectorized by Doc2Vec, classified into multiple clusters by the k-means method, and quantified to be used as explanatory variables for predicting traffic accidents and pre-dicting fatalities, and serious injuries. As a result, the possibility of confirming the characteristics of traffic accidents concerning the elderly was obtained, and the possibility of improving the prediction accuracy by using the accident summaries as the explanatory variable was confirmed.