Feature selection to predict COVID-19 new patients in the four southern border provinces of Thailand

Abstract

This paper presents a machine learning-based prediction framework that utilizes ensemble feature selection techniques to accurately forecast the number of new coronavirus disease (COVID-19) infections in Thailand’s four southern border provinces. The framework used include multiple linear regression (MLR), mul tilayer perceptron neural networks (MLP-NN), and support vector regression (SVR), to classify short-term trends in new patient cases. The study evaluates the effectiveness of these models across different provinces and demonstrates how integrating feature selection methods: forward selection (FS), backward elimination (BE), and genetic algorithms (GA) enhances prediction accuracy. The findings highlight the adaptability of the models, with each province ben efiting from tailored model-feature selection strategies. The results show that the predictive models align closely with real patient data, enabling authorities to anticipate outbreaks and implement timely interventions. Moreover, the pro posed methodology can be applied to other epidemics, making it a valuable tool for public health planning and preparedness. The study offers actionable in sights for decision-makers, emphasizing the importance of predictive modeling in community-level outbreak management.