AI decodes what really causes cars to pollute and how to stop it

While hybrids significantly reduce emissions in urban and low-speed driving conditions, the authors highlight that their environmental advantage diminishes during high-load operations, when combustion assistance from the engine increases. This finding underscores the need for continuous improvement in hybrid technologies, particularly in optimizing power distribution between electric and combustion components under varying driving conditions.

A new study introduces a breakthrough in sustainable transport modeling through the use of explainable machine learning. The research, titled “Explainable Machine Learning Prediction of Vehicle CO₂ Emissions for Sustainable Energy and Transport,” published in Energies, presents a comparative framework that can predict carbon dioxide (CO₂) emissions from internal combustion engine (ICE) and hybrid electric vehicles (HEVs) with unprecedented accuracy.

The study, based on certified data from the UK Vehicle Certification Agency (VCA), integrates predictive modeling with transparency-driven artificial intelligence, marking a major step toward making vehicle emission analysis interpretable and policy-relevant. Using XGBoost, a high-precision machine learning algorithm, the researchers achieved R² values above 0.98, revealing that explainable AI can accurately forecast vehicle emissions while clarifying the underlying technical and environmental drivers.

Understanding what drives vehicle emissions

The research addresses one of the most persistent challenges in sustainable transportation: identifying which mechanical and operational factors most influence CO₂ output across different vehicle types. The team applied advanced predictive models to 2,000 certified vehicles, analyzing over 20 key variables including engine displacement, fuel consumption, noise levels, and pollutant emissions.

The results confirmed that engine capacity and medium-speed fuel consumption are the dominant contributors to CO₂ emissions in internal combustion vehicles. In contrast, for hybrid electric vehicles, the study found that high-load fuel consumption and nitrogen oxide (NOₓ) emissions were the strongest predictors of carbon output.

While hybrids significantly reduce emissions in urban and low-speed driving conditions, the authors highlight that their environmental advantage diminishes during high-load operations, when combustion assistance from the engine increases. This finding underscores the need for continuous improvement in hybrid technologies, particularly in optimizing power distribution between electric and combustion components under varying driving conditions.

A major innovation of the study lies in the use of Accumulated Local Effect (ALE) analysis, an explainability technique that reveals how each variable interacts with others to influence emissions. Unlike traditional black-box AI models, ALE provides transparent cause-and-effect insights, allowing policymakers and engineers to understand not just how much emissions a vehicle produces, but why.

Noise levels and engine load: A hidden link to emissions

Beyond traditional factors, the research introduces a novel environmental dimension by incorporating vehicle noise levels as an explanatory variable for emissions. Noise, measured in decibels (dB), emerged as a meaningful indicator of engine stress and driving intensity, both of which correlate strongly with CO₂ output.

The analysis revealed a distinct threshold effect: once noise levels exceed 60 dB(A), emissions increase sharply. This correlation reflects how aggressive driving, poor road surfaces, or high engine load amplify both acoustic and environmental pollution. The study thereby connects acoustic emissions and air pollution, two typically separate fields, into a unified sustainability model.

By linking these parameters, the research opens new possibilities for dual-impact environmental policies, where reducing traffic noise could simultaneously lower carbon emissions. This insight carries important implications for urban sustainability and transport planning, as cities look to integrate energy, air quality, and acoustic data in smart mobility systems.

Moreover, the findings reinforce that vehicle design, material efficiency, and tire–road interactions can indirectly influence both sound and emission profiles, emphasizing the interconnected nature of transport sustainability.

The researchers also demonstrate that hybrid systems, while quieter at lower speeds, experience emission surges when combustion engines activate under high-load driving, such as rapid acceleration or steep terrain. This dynamic highlights the conditional nature of hybrid vehicle efficiency and underscores the continued importance of vehicle electrification and adaptive driving strategies in mitigating emissions.

AI for transparent and sustainable policy design

Perhaps the most impactful aspect of the study is its focus on explainable artificial intelligence (XAI), a branch of machine learning that prioritizes interpretability alongside accuracy. In traditional emission models, even highly predictive algorithms act as “black boxes,” providing numerical results without clarity on how each input influences the outcome.

On the other hand, the XGBoost model developed by Yuan and colleagues integrates interpretability directly into its framework. Through Accumulated Local Effect plots, the model identifies both direct and conditional relationships among variables, allowing for real-time visualization of how changes in engine capacity, fuel economy, or acoustic intensity affect emissions.

This transparency has significant implications for policy development and vehicle regulation. Regulators can now pinpoint which parameters yield the highest reductions per unit of technological investment, whether through downsizing engines, improving mid-speed fuel efficiency, or mitigating NOₓ under load.

The study’s insights support a more differentiated approach to vehicle policy:

• For internal combustion engines, priority measures include downsizing engine capacity, improving fuel efficiency in medium-to-high-speed conditions, and promoting lighter materials to reduce load.
• For hybrid vehicles, strategies should focus on extending electric driving ranges, optimizing powertrain control under high-demand conditions, and combining dual CO₂–NOₓ mitigation technologies, such as selective catalytic reduction (SCR) and exhaust gas recirculation (EGR).
• Urban noise reduction policies, including smoother road design, improved vehicle insulation, and acoustic emissions standards, can produce synergistic reductions in both CO₂ and noise pollution.

By bridging data-driven prediction and human-understandable explanation, the research provides governments, automakers, and energy planners with a practical decision-support tool for accelerating decarbonization.

Toward smarter and cleaner mobility

Hybrid electric vehicles remain a vital transitional technology in the shift toward zero-emission transport but cautions that their environmental benefits are not absolute. Hybrid systems perform best under balanced energy loads but lose efficiency when reliance on combustion engines rises, an important consideration for both vehicle design and consumer education.

According to the authors, integrating explainable AI into environmental modeling not only improves predictive accuracy but also enhances trust and accountability in policymaking. As nations move toward carbon neutrality, the ability to trace emissions back to their technical causes will be essential in designing equitable, evidence-based climate policies.

The introduction of noise as an environmental co-variable is especially notable. By quantifying how engine acoustics correlate with emission intensity, the study creates a multi-dimensional sustainability framework, linking transportation technology with environmental psychology and public health.

The findings align with Sustainable Development Goal 13 (Climate Action) and the broader agenda for low-carbon, data-driven transport ecosystems. They illustrate how machine learning, when made transparent and interpretable, can become a cornerstone of sustainable engineering.

The researchers propose future research focused on integrating electric vehicle (EV) and hybrid fleet data into similar models to further refine emission forecasting and support the global transition to net-zero transport systems.