Meta-learning applied to a multivariate single-step fusion model for greenhouse gas emission forecasting in Brazil

Abstract

Climate change, driven by greenhouse gas (GHG) emissions, causes extreme weather events, impacting ecosystems, biodiversity, population health, and the economy. Predicting GHG emissions is crucial for mitigating these impacts and planning sustainable policies. This research proposes a novel machine learning model for GHG emission forecasting. Our model, named the meta-learning applied to multivariate single-step fusion model, utilizes historical GHG data from Brazil over the past 60 years. It predicts multivariate time series, meaning it can consider multiple factors simultaneously, leading to more accurate forecasts. Additionally, the model employs two innovative techniques: (i) fusion model aligns different data sources to ensure compatibility and improve prediction accuracy and (ii) meta-learning allows the model to learn from past prediction tasks, generalizing better to new data and reducing the need for large training datasets. Compared to the widely used Bidirectional Long Short-Term Memory (BiLSTM) model, our approach achieves significantly better results. On the same dataset, it reduces the mean absolute percentage error by 116.84% with 95% confidence, demonstrating its superior performance. Furthermore, the model's flexibility allows it to be adapted for predicting other multivariate substances, making it a valuable tool for various environmental studies.