Machine Learning to Predict Energy Usage Outcomes

Abstract

This study evaluates the practicality of global sustainability targets by predicting renewable
energy share and CO2 emissions based on a range of socio-economic, environmental and
geographical factors. Using country-level data from 2000-2020, we developed a machine
learning model to estimate the renewable energy percentage within total energy consumption
and forecast CO2 emissions. We trained a multi-layer perceptron (MLP) regressor on a dataset
sourced from the World Bank and International Energy Agency, achieving an r-squared value of
0.94 for renewable energy share predictions and 0.99 for CO2 emissions predictions. These high
accuracies suggest that this model could support policymakers in setting achievable
sustainability goals tailored to specific national circumstances.