환경 요인과 계절성을 통합한 머신러닝 기반 쯔쯔가무시병 발생 예측 모델 개발

Integrating Environmental and Seasonal Factors in a Machine Learning Model for Predicting Scrub Typhus Incidence

Background: Scrub typhus poses a significant public health threat in endemic regions as a vector-borne infectious disease. While accurate prediction of this disease is crucial for effective prevention and control, conventional models often fail to fully capture the complex nonlinear interactions among seasonal patterns, ecological mechanisms, and environmental factors. Objectives: We aimed to develop predictive models that integrate seasonal and environmental factors to accurately forecast scrub typhus incidence and provide ecological insights into its transmission dynamics. Methods: Three models were constructed and compared using monthly scrub typhus case data and meteorological data from 2013 to 2022. A SARIMA model was applied to analyze seasonal patterns, an ARIMAX model was used to incorporate environmental factors, and Random Forest-based regression and classification models were employed to analyze nonlinear relationships and complex interactions among variables. Results: The Random Forest classification model demonstrated superior predictive performance with an AUC of 0.955 and MCC of 0.591. Partial dependence analysis revealed that ‘case count from 12 months prior’ and ‘temperature from three months prior’ were the most influential predictors. Temperature from three months prior showed a distinct negative correlation with outbreak risk within the range of 5.9~16.8°C, while cases from 12 months prior exhibited a nonlinear relationship with peak risk observed within a specific range (3.0~14.4 cases). Conclusions: Machine learning approaches combined with partial dependence analysis effectively captured and interpreted the complex ecological mechanisms underlying scrub typhus incidence. The models developed in this study can serve as a scientific basis for establishing early warning systems and formulating targeted prevention strategies.