Truace
Subscribe

Machine learning and SHAP-based explainability for accident severity prediction: evidence from a strategic road network in a developing country

Objective This study aims to develop an Explainable Artificial Intelligence (XAI) framework to classify and predict road accident severity on a strategic road network in a developing country. The research focuses on identifying the primary risk factors contributing to fatal and serious injuries to support "Vision Zero" policies. Methods Three machine learning algorithms-Random Forest (RF), XGBoost, and Artificial Neural Networks (ANN)-were trained using a dataset of traffic accidents from federal highways in Per…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-1771· cite & verify
Machine learning and SHAP-based explainability for accident severity prediction: evidence from a strategic road network in a developing country
"M67 motorway traffic jam - geograph.org.uk - 3415782" by Steve Fareham is licensed under CC BY-SA 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by-sa/2.0/.

In brief

Researchers built an explainable AI framework to predict how severe road accidents are on federal highways in Pernambuco, Brazil. They compared Random Forest, XGBoost and neural network models, and tested whether artificially rebalancing the data with SMOTE helped.

The work matters because it turns opaque prediction models into findings that road authorities could use to target infrastructure upgrades and enforcement under Vision Zero goals. The abstract reports no real-world deployment or measured reduction in crash deaths, so the safety benefit remains untested.

Main points

  1. Three algorithms were trained on traffic accident data from federal highways in Pernambuco, Brazil.
  2. The Random Forest on the original data distribution performed best, with an AUC-ROC of 0.952 and an F1-Score of 0.747.
  3. SHAP analysis tied fatal outcomes to the combination of single-lane roads, head-on collisions and excessive speed.
  4. The authors conclude that keeping the original data distribution is more reliable than artificial balancing.

The gain

A Random Forest model trained on unbalanced highway crash data predicted accident severity with high accuracy, and SHAP analysis exposed the main drivers of lethality for engineers and policymakers.

The rundown

The study aims to support Vision Zero policies by identifying the main risk factors behind fatal and serious injuries on a strategic road network in a developing country.

To handle imbalance between severe and non-severe crashes, the team compared models on the original data against data oversampled with the Synthetic Minority Over-sampling Technique. The original distribution gave the stronger and more methodologically sound results.

SHapley Additive exPlanations were used to interpret the models' decision-making, so that complex black box predictions could be decoded into factors engineers and policymakers can act on in emerging economies.

Sources

  1. Peer-reviewedTraffic Injury Prevention2026-10-07

The debate