Predictive Modeling of Carbon Dioxide Emissions from Gasoline-Powered Vehicles Using an Artificial Neural Network

This study presents a predictive modeling to estimate carbon dioxide emissions from gasoline-powered vehicles based on the engine size and fuel consumptions. The predictive model was built using the Multi-Layer Perceptron (MPL) as a machine learning method. First, a concise analysis of the dataset was conducted using the RapidMiner open-source software. Furthermore, the neural network architecture and the key hyperparameters optimization were outlined. Subsequently, the results obtained by this model were presented and discussed. The optimization method, based on the Grid Search technique, revealed that for the present case, the optimal values for the learning rate and momentum were 0.01 and 0.8, respectively. With these optimized parameters, the performance of the Multi-Layer Perceptron was improved. The model displayed then a good accuracy when predicting CO
2
values with a new unseen dataset, with a root mean square error of 2.531 and a coefficient of determination of 0.999.