Carbon Emission Prediction Through the Harmonization of Extreme Learning Machine and INFO Algorithm

Abstract

This research introduces a novel optimization algorithm, weIghted meaN oF vectOrs (INFO), integrated with the Extreme Learning Machine (ELM) to enhance the predictive capabilities of the model for carbon dioxide (CO2) emissions. INFO optimizes ELM's weight and bias. In six classic test problems and CEC 2019 functions, INFO demonstrated notable strengths in achieving optimal solutions for various functions. The proposed hybrid model, ELM-INFO, exhibits superior performance in forecasting CO2 emissions, as substantiated by rigorous evaluation metrics. Notably, it achieves a superior R2 value of 0.9742, alongside minimal values in Root Mean Squared Error (RMSE) at 0.01937, Mean Squared Error (MSE) at 0.00037, Mean Absolute Error (MAE) at 0.0136, and Mean Absolute Percentage Error (MAPE) at 0.0060. These outcomes underscore the robustness of ELM-INFO in accurately predicting CO2 emissions within the testing dataset. Additionally, economic growth is the most significant element, as indicated by ELM-INFO's permutation significance analysis, which causes the model's MSE to increase by 19%. Trade openness and technological innovation come next, each adding 7.6% and 8.1% to the model's MSE increase, respectively. According to ELM-INFO's performance, it's a powerful tool for developing ecologically sound policies that improve environmental resilience and sustainability. © 2013 IEEE.