An Ensemble Method for Early Prediction of Dengue Outbreak
提出一种结合负二项回归、ARIMA和GLARMA模型的集成方法，利用地形和气候协变量的滞后值，在圣胡安和伊基托斯的数据上实现提前8周预测，平均绝对误差小于10例。
Abstract Predicting a dengue outbreak well ahead of time is of immense importance to healthcare personnel. In this study, an ensemble method based on three different types of models has been developed. The proposed approach combines negative binomial regression, autoregressive integrated moving average model and generalized linear autoregressive moving average model through a vector autoregressive structure. Lagged values of terrain and climate covariates are used as regressors. Real-life application using data from San Juan and Iquitos shows that the proposed method usually incurs a mean absolute error of less than 10 cases when the predictions are made 8 weeks in advance. Furthermore, using model confidence set procedure, it is also shown that the proposed method always outperforms other candidate models in providing early prediction for a dengue epidemic.