Our paper Highway to… Determining Fatal Outcomes in Traffic Accidents Based on Police Reports got accepted for publication at the 35th Brazilian Conference on Intelligent Systems (BRACIS).

The paper addresses Brazil’s ongoing road safety challenges, particularly the high number of fatal accidents on its federal highways. With one of the largest road networks in the world, the country faces significant social and economic impacts from traffic-related incidents. This study presents a case analysis focused on the southern region of Brazil, applying and comparing three machine learning models — Random Forest (RF), k-Nearest Neighbors (kNN), and Multilayer Perceptron (MLP) — to classify the severity of road accidents.

Using open data provided by the Brazilian Federal Highway Police (PRF) from 2021 to 2024, extensive preprocessing was conducted, including categorical variable encoding, feature selection, and the application of the SMOTE technique to mitigate class imbalance. The models were evaluated using key performance metrics such as specificity, F1-score, and AUC-ROC. Among the findings, both RF and kNN (with SMOTE) demonstrated excellent performance in predicting fatal accidents, each achieving an AUC-ROC of 0.99. In addition to model evaluation, the paper includes a post-hoc analysis using Shapley Additive Explanations (SHAP) to identify the most influential features associated with accident severity, providing valuable insights into the factors that contribute most to fatalities on Brazilian highways.

As detailed in the conference website: “BRACIS is one of the most important events in Brazil for researchers aiming to publish significant and novel results in the fields of Artificial Intelligence (AI) and Computational Intelligence (CI). It was established through the merger of the two most prominent scientific events in Brazil dedicated to AI (SBIA) and CI (SBRN).”

As soon as published, the paper will be linked at the Publications page.