Building a Global CO₂ Emissions Estimator Using Machine Learning

ELEKS’ Data Science team built a global CO₂ emission estimator, which can forecast emissions for the following year using natural Machine Learning approaches. Predictions are modelled from data taken from the World DataBank, which allows users to access various global datasets.

The team opted to include estimates for all four CO₂-related targets; total CO₂ emissions, solid fuel CO₂ emissions, liquid fuel CO₂ emissions, and gaseous fuel CO₂ emissions, using a top-down approach and estimating the component parts based on total CO₂ emissions.

Once they had the data, the experts filtered it to give the model the most accurate outcomes. Then, the team applied several strategies to reduce the dimension of the matrix and refine obtained results by removing collinear features.

They evaluated multiple models and algorithms to estimate the CO₂ emissions and carried out model validation to show those that worked best. Our experts then applied different strategies and techniques to improve the precision of results. They evaluated the model with the multirun holdout validation, calculating the variance and concluding on the robustness of the model.