Explainable support vector regression coupled with quantum firefly optimisation algorithm for carbon emission prediction in West Africa: The role of socioeconomic, energy, and environmental factors

Achieving the Sustainable Development Goals and creating sustainable environmental policies depend on accurate carbon emission prediction. The Autoregressive model (AR), Autoregressive Integrated Moving Average (ARIMA), Multiple Linear Regression (MLR), Extreme Learning Machine (ELM), Support Vector Regression (SVR), ELM optimised by Quantum Firefly Optimisation Algorithm (ELM-QFOA), and SVR combined with the same algorithm (SVR-QFOA) are the models evaluated in this study for predicting carbon emissions in West Africa. The best prediction accuracy is demonstrated by the SVR-QFOA model. According to the training sample, the SVR-QFOA model improved the prediction accuracy of AR by 95.37 % for Côte d’Ivoire, 92.31 % for Ghana, 96.85 % for Nigeria, and 78.81 % for Senegal. For Côte d’Ivoire, Ghana, Nigeria, and Senegal, it enhanced AR performance by 96.43 %, 95.35 %, 97.77 %, and, 83.07 % for Côte d’Ivoire, Ghana, Nigeria, and Senegal, respectively, in the testing sample. According to the study, the top models are SVR-QFOA for Ghana and Nigeria, SVR for Côte d’Ivoire, and ELM for Senegal. The study employs Shapley Additive Explanations (SHAP) to quantify the contribution of each factor to emissions prediction. The study recommends the use of SVR-QPSO, SVR and ELM for climate policies aimed at improving environmental resilience in West Africa.