Data-Driven Models Support a Vision for Over-the-Air Vehicle Emission Inspections

Emissions inspection and maintenance (I/M) programs for light-duty motor vehicles manage ambient air quality by enforcing emissions standards and requiring non-compliant vehicles to be repaired or retired. I/M programs in the United States typically identify over-emitters through on-board diagnostics (OBD) systems and vehicles’ proprietary firmware (i.e., indirect tests), rather than through physical measurements of exhaust gases (i.e., tailpipe tests). Analyzing data from Colorado’s I/M program, this study finds the OBD test to have an accuracy of 87%, but a false pass rate of 50%, when predicting the result of a corresponding tailpipe test. As an alternative, transparent data-driven models—using logistic regression and gradient boosting machines—to consistently identify over-emitting vehicles are proposed. These models were up to 24% more accurate, or 85% more sensitive than the current OBD test in a stratified data sample. A key benefit of transparent statistical models—jurisdictions’ ability to tune the test methods to best suit program needs—is also assessed. Finally, this study shows how these results support a vision for cloud-based, selective I/M programs where statistical models are applied to OBD data—collected over-the-air from vehicles—to identify and require additional inspection for only the most probable over-emitters.