Optimising SVR for epidemiological predictions: a case study on COVID-19 mortality in Japan

This study enhances support vector regression machine (SVR) for COVID-19 mortality forecasting in Japan using three particle swarm optimisation (PSO) variants. Our main contributions include: 1) achieving superior model performance, notably with the fast convergence PSO-SVR variant, which outperforms existing models with an R-Squared value of 0.717; 2) demonstrating consistent and improved prediction accuracy across various PSO variants; 3) establishing the potential of our methods for broader applications beyond epidemiological modelling. Our findings, significantly advancing the accuracy and efficiency of predictive analytics in this domain, are benchmarked against prior studies, showing notable improvements in SVR hyperparameter optimisation.