IMPROVING NOROVIRUS OUTBREAK PREDICTION THROUGH FEATURE SELECTION USING MACHINE LEARNING METHODS

초록

Norovirus disease in the Republic of Korea consistently displays a seasonal pattern, especially prominent during the winter months. Although the timing of outbreak periods can change due to climate change, these outbreaks continue to occur regularly. It is crucial to analyze these seasonal patterns to predict the occurrence of the outbreak. Notably, no quantitative criteria have been established to determine the onset of norovirus outbreaks. In this study, we aim to establish criteria for predicting the occurrence of norovirus outbreaks using various machine learning classification methods, including Support Vector Machines (SVM), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and a Long Short-Term Memory (LSTM). By utilizing meteorological data, past incidence data, and other variables, we predict norovirus outbreak periods based on the criteria suggested in the present study. The training period spans from 2016 to 2022, while the 2023 data is used for testing. We compare the prediction accuracy of outbreak periods based on the selected variables. Our results indicate that among the machine learning methods tested, SVM and RF achieved the highest F1-scores in classifying outbreak periods. Additionally, the models accurately predicted the first thirteen weeks of 2023 as an outbreak period. By accurately predicting the occurrence of norovirus outbreaks, our approach enables the early detection of an increase in the number of norovirus cases, allowing for proactive measures and prevention strategies to mitigate the impact of these outbreaks.

키워드