Conference Paper

Enhanced Support Vector Regression for Accurate CO2 Emission Prediction from

Accurate 〖CO〗_2 Emission prediction is essential for evidence-based climate policy, yet Support Vector Regression (SVR), a leading machine learning approach for prediction tasks, requires precise hyperparameter selection, which standard optimizers often address poorly due to premature convergence. This chapter proposes SCMSSA, a three-mechanism enhancement of the Salp Swarm Algorithm (SSA) that integrates logistic-chaotic single-dimensional perturbation, convex lens mirror imaging opposition, and sine-cosine follower perturbation to simultaneously improve exploration coverage, population diversity, and convergence speed, thereby optimizing the parameters of SVR. Applied to monthly U.S. Energy Information Administration records (1973-2023), the SVR-SCMSSA model attains an accuracy of 95.78%, surpassing all compared models on every evaluation metric. Keywords: Salp Swarm Algorithm (SSA), Support Vector Regression (SVR), Energy Consumption, United States, Carbon Emission