A novel hybrid machine learning model for prediction of CO2 using socio-economic and energy attributes for climate change monitoring and mitigation policies

Abstract

Industrial development has contributed to carbon emissions majorly, resulting in high concentrations of greenhouse gases (GHGs) in the environment leading to climate change phenomenon. Climate change threatens people in multiple ways: threatening food security, water scarcity, frequent and extreme weather events, the spread of diseases, economic losses and migration etc. The World Health Organization (WHO) declared that climate change is the greatest threat to global health in the 21st century. Since 1970, CO2 emission has increased by about 90 per cent, making it 78 per cent of the total greenhouse gas emission. Climate change impact, carbon emission factors and social-economic attributes make the prediction of GHG emission a very complex research problem having dynamic scenarios due to a large number of factors and impacting variables. Accurate prediction of carbon emission in such a scenario makes it one of the most important and challenging research works. Artificial intelligence and machine learning approaches are increasingly being used to study complex, dynamic environmental phenomenons with high variability of time, space and other factors. The research paper proposes hybrid machine learning models for the prediction of CO2 emissions using energy and social-economic variables. The work uses energy and socioeconomic variables from 1960 to 2018 to collate them to provide a new perspective on the application of machine learning approaches in the modelling and prediction of GHG emissions. The proposed hybrid model of principal component analysis (PCA) and machine learning approaches is compared on accuracy and efficiency and performs better than other machine learning and deep learning approaches such as Linear regression variants, random forest regression (RFR), support vector regression (SVR), recurrent neural network (RNN), long short term Memory (LSTM) and Tabnet etc. The proposed hybrid model reports MAE equal to 0.0307, RMSE equal to 0.0346, MAPE equals to 5.1447 and SMAPE equal to 5.2267. In terms of efficiency of computation, the proposed hybrid machine learning model PCA+LER took 12.4 ms better than other models. This proposed research work of correlation analysis and prediction can help policymakers and governments in the mitigation and management of carbon emissions.