Deep learning model based prediction of vehicle CO₂ emissions with eXplainable AI integration for sustainable...

Area of Science:

• Environmental Science
• Computer Science
• Automotive Engineering

Background:

• Transportation sector is a major contributor to climate change via carbon dioxide (CO2) emissions.
• Vehicle emissions exacerbate global warming and extreme weather events.
• Predicting vehicle CO2 emissions is crucial for developing mitigation strategies.

Purpose of the Study:

• To develop and validate a deep learning model for predicting vehicle CO2 emissions.
• To enhance model interpretability using eXplainable Artificial Intelligence (XAI).
• To identify key vehicle attributes influencing CO2 emissions.

Main Methods:

• Utilized a Canadian government open data portal dataset.
• Employed deep learning, specifically a Multilayer Perceptron (MLP) architecture named CarbonMLP.
• Integrated eXplainable Artificial Intelligence (XAI) methods, including SHapley Additive exPlanations (SHAP).
• Performed hyperparameter tuning for model optimization.

Main Results:

• The CarbonMLP model achieved high accuracy with R-squared of 0.9938 and MSE of 0.0002.
• High-performance engines and fuel consumption (city/highway) were identified as significant contributors to emissions.
• SHAP analysis provided insights into feature importance for emission prediction.
• Identified skewed distributions in manufacturer vehicle production and fuel type consumption trends.

Conclusions:

• The proposed deep learning and XAI methodology accurately predicts vehicle CO2 emissions.
• The findings support the development of targeted strategies for emission reduction.
• Further research is recommended to expand the dataset, include other pollutants, and explore real-world applications.