ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING IN RISK MANAGEMENT IN THE ROAD TRANSPORT OF HAZARDOUS PRODUCTS IN BRAZIL: FUNDAMENTALS, APPLICATION, AND COMPLIANCE

Abstract

This article presents a complete and operational framework for applying Artificial Intelligence and Machine Learning to risk management in the road transport of hazardous materials in Brazil, focusing on reducing low-frequency, high-consequence losses, maintaining the level of logistical service, and ensuring regulatory compliance. The work aligns with Decree No. 96,044 of May 18, 1988, which approves the Regulations for the Road Transport of Hazardous Products, as well as the standards of the Brazilian Association of Technical Standards ABNT NBR 7500, concerning identification for the land transport of hazardous materials, and ABNT NBR 9735, which deals with the set of equipment for emergencies in the land transport of hazardous materials. Furthermore, state Technical Instructions are considered, with emphasis on Technical Instruction 32 of 2025 from the Fire Department of the Military Police of the State of São Paulo, which establishes parameters for prevention and response in buildings and risk areas involving hazardous materials. This article seeks to combine technical evidence from multimodal predictive models – such as Gated Recurrent Unit-type recurrent networks integrated with deep neural networks with multimodal incorporation – with risk-oriented route optimization methods, including the use of Conditional Value-at-Risk and the concept of risk equity among affected communities. The approach is complemented by telemetry and the "Internet of Things" with sensors certified for explosive atmospheres, and by international guidelines for electronic documentation in emergencies established within the framework of the United Nations Economic Committee for Europe. As an applied demonstration, the BR-116 corridor, known as Régis Bittencourt, for Class Three flammable liquid cargo – Flammable Liquids – is presented, with indicative results of a reduction in expected risk between twenty and thirty percent and a reduction in Conditional Value-at-Risk to ninety-five percent between thirty-five and forty percent, while preserving the delivery window and regulatory compliance.