A Bayesian Network for the Analysis of Traffic Accidents in Peru

Traffic accidents are a problem that affects the State and society, because they cause material damage, injuries and even the death of a person. This has led countries such as China, Switzerland and Australia to carry out studies using Bayesian networks to determine the main causes and, based on them, propose measures to reduce the number of traffic accidents. Following this trend, we, without having any expert knowledge on the subject, decided to analyze the data of traffic accidents on the Pan-American Highway in Lima, Peru. This analysis was done by means of directed graph learning with the Hill Climbing Search, Chow-Liu, K2, BIC and BDEU. In addition, we used a Bayesian estimator to calculate the conditional probability distribution for our dataset. This dataset contains observations from the years 2017 to 2019 and approximately 16 km of this highway. Our results show that it is possible to identify the possible causes of excess accidents in specific areas of the Pan-American Highway in certain shifts i.e., 32% of fatal accidents occur between 12 am and 7 pm in the Rimac district and of these 20% are due to pedestrians on the highway.