From Predictive Accuracy to Human-Centric Decision Support: An Operational HCT-ML Framework for Intelligent Transportation Systems

Résumé fourni par la source

Machine learning in intelligent transportation systems (ITS) is commonly evaluated through predictive performance, yet deployment decisions also depend on whether outputs are relevant to a defined decision, understandable to intended users, equitable across affected groups, uncertainty-aware, subject to appropriate human authority, and actionable within operational constraints. This Perspective develops the Human-Centric and Trustworthy Machine Learning (HCT-ML) framework as an operational decision-support profile for ITS. Its novelty is not the invention of new responsible artificial intelligence (AI) principles; rather, HCT-ML integrates established requirements around four transport-specific constructs—decision owner, decision horizon, intervention pathway, and consequence of error—and translates six human-centric dimensions (relevance, explainability, fairness, uncertainty, human oversight, and actionability) into evidence requirements, candidate indicators, context-specific thresholds, and non-compensatory deployment gates. The framework is benchmarked against the National Institute of Standards and Technology (NIST) AI Risk Management Framework, Organisation for Economic Co-operation and Development (OECD) AI Principles, the European Union (EU) AI Act, Institute of Electrical and Electronics Engineers (IEEE) 7000-series standards, and United States Department of Transportation (USDOT) AI-assurance guidance. We further provide a formal operationalization, application-specific priority profiles, real-world safety cases from automated-driving investigations, and a worked municipal road-safety protocol using the Montréal open-data context. The article does not claim empirical validation or causal safety gains; instead, it provides an evidence-informed and reproducible protocol that can be tested through stakeholder studies, simulations, and field deployments.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.