Beyond Regulatory Thresholds: Continuous and Explainable Assessment of Vehicle Emission Condition

Vehicle emission testing commonly evaluates regulatory compliance using predefined thresholds and pass/fail outcomes, providing limited information about differences in multivariate emission condition. This study develops a vehicle emission assessment framework combining hierarchical machine learning and explainable artificial intelligence to provide continuous and interpretable information alongside regulatory assessment. The Engine Freshness Index (EFI) represents vehicle emission condition relative to a data-derived clean-combustion reference state through emission-pattern discovery and distance-based modeling. A constrained Random Forest model learned this relationship for new inspection records, achieving a coefficient of determination (R²) of 0.984. Explainability analysis showed that predictions were driven by physically interpretable emission characteristics. Structured diagnostic reasoning and large language model-assisted interpretation translated analytical outputs into decision-support information, with the selected language model achieving an evaluation score of 94.8%. The framework demonstrates how routinely collected inspection measurements can complement regulatory thresholds through continuous assessment of vehicle emission condition. Preprint. This manuscript has been submitted to Transportation Research Part D: Transport and Environment and is currently under journal consideration.