Net-zero futures cities and transportation systems: estimation and analyzing of vehicle’s carbon dioxide production by knowledge transferring

The limited energy resources, critical climate change conditions, and globalwarming, coupled with today’s enormous industrial
development, necessitate innovative approaches to control the situation. The automotive industry and its pollution emissions
remain among the top environmental concerns. In this article, we present a progressive plan that leverages deep neural
networks and inductive transfer learning methods to develop a CO2 forecasting application while addressing the challenge
of limited dataset size on a large-scale problem. Two datasets regarding vehicle emission productions in Europe (source
dataset) and Canada (target dataset) with completely different statistic properties are used to demonstrate this approach. The
proposed method shows promising results with the effectiveness of the inductive transfer learning model. It demonstrates the improvement of model generalization despite limited data availability and huge data characteristic differences. The RMSE of the model on the European dataset is 20.33 when implementing a traditional deep neural network. For the Canadian dataset, using the inductive transfer learning method reduces the RMSE by 6.29 compared to the traditional machine learning approach. The proposed work provides great insight for policymakers on the future of the automotive industry to promote new solutions and technological advancements and also for consumers to make better decisions on their future choices. The results are qualitatively reported and discussed.