Prediction of greenhouse gas (GHG) emissions is vital to minimize their negative impact on climate change and global warming. In this thesis, we propose new models based on data mining and supervised machine learning algorithms (Regression and classification) for predicting GHG emissions arising from passenger and freight road transport in Canada. Four categories of models are investigated namely artificial neural network multilayer perceptron, multiple linear regression, multinomial logistic regression and decision tree models. From the application results, it was found that artificial neural network multilayer perceptron model showed better predictive performance over other models. Ensemble technique (Bagging & Boosting) was applied on the developed Multilayer Perceptron model which significantly improved the model's predictive performance.
The independent variable importance analysis conducted on multilayer perceptron model disclosed that among the input attributes Light truck emissions, Car emissions, GDP transportation, Heavy truck emission, Light duty truck fuel efficiency, Interest rate (overnight), Medium Trucks Emission, Passenger car fuel efficiency and Gasoline Price have higher sensitivity on the output of the predictive model of GHG emissions by Canadian road transport.
Scenario analysis is conducted using widely available socioeconomic, emission and fuel efficiency attributes as inputs in multilayer perceptron (with bagging) model. The results show that in all Canadian road transport GHG emission projection scenarios, all the way through 2030, emissions from Light trucks will hold a major share of GHG emissions. Thereby, rigorous efforts should be made in mitigating GHG emissions from these trucks (freight transport) to meet the ambitious GHG emission target for Canadian road transport.