Framework For Vehicle Environmental Impact: Machine Learning Approach In Canadian Policy

Abstract

In this dissertation, we investigate using machine learning methodologies to categorize vehicles based on their ecological impact in the Canadian context, focusing on vehicle emissions and their influence on climate change and public health. The study commences with identifying key variables affecting vehicle emissions, such as CO2 emissions, fuel efficiency, engine size, fuel type, vehicle weight, and age. The research establishes strong correlations among these variables through comprehensive exploratory data analysis, particularly highlighting the inverse relationship between fuel efficiency and CO2 emissions and the higher emissions associated with more extensive and older vehicles. These findings underscore the importance of targeted policies to advocate for fuel-efficient, smaller, and newer vehicles to mitigate emissions.

The research further examines the relationships among fuel efficiency, fuel type, and vehicle emissions, confirming that fuel-efficient vehicles consistently produce lower emissions. Electric and hybrid vehicles, in particular, significantly outperform gasoline and diesel vehicles in terms of environmental impact. This analysis emphasizes the environmental benefits of transitioning to electric and hybrid technologies and lays the groundwork for policy recommendations to support this transition through incentives and infrastructure development.
A central focus of the study is developing and evaluating machine learning models for classifying vehicles based on their environmental impact. Various algorithms, including Decision Trees, Random Forests, Gradient Boosting Machines (GBM), Support Vector Machines (SVM), and K-nearest neighbours (KNN), are assessed for their effectiveness. The GBM algorithm is identified as the most accurate and robust, positioning it as a crucial tool for policymakers in identifying high-impact vehicles and designing targeted regulations. Despite challenges such as data quality and model interpretability, the success of the GBM model demonstrates the potential of machine learning in driving data-informed policy decisions and promoting sustainable transportation.
To conclude, this dissertation highlights the capability of machine learning to comprehend and diminish the environmental impact of vehicles. The research provides practical insights that can significantly enrich transportation policies by identifying critical variables, scrutinizing their interrelationships, and devising effective classification models. These insights can further support the efforts of the Canadian government in reducing greenhouse gas emissions and promoting sustainable development. Future research is suggested to integrate real-time data, broaden the geographic and temporal scope, explore advanced machine learning techniques, enhance model interpretability, and formulate scalable solutions for broader implementation.