Weekly forecasting model for dengue hemorrhagic fever outbreak in Thailand

Abstract

A dengue virus causes diseases, including dengue hemorrhagic fever (DHF) which induces several sicknesses and deaths in Thailand. DHF is categorized as one of the most dangerous communicable diseases by the Ministry of Public Health Thailand (MoPH); moreover, the MoPH also sets strict protocols and encourages forecasting techniques for monitoring and dealing with the outbreaks. This research aims to utilize the data that were gathered from external sources, e.g. Google Trends data and meteorology data, to forecast the number of cases that will occur within the 7 day-interval in the next 1–4 weeks. Six provinces—including Chiang Rai, Mukdahan, Pattani, Phichit, Ayutthaya, and Ratchaburi—were selected as they represent the unique patterns of dengue outbreaks in Thailand. The machine learning models—including Random Forest, AdaBoost, Extra-Trees, and Regularized Regressions—were used to forecast the number of the cases. The performances of these models were compared to the performances of the traditional time series model including Naïve model and Moving Average. The proposed machine learning models for Chiang Rai, Mukdahan, and Pattani yield better results than those of the traditional models.