A Rule-Based Ensemble Approach for Diarrhoea Outbreak Prediction

Abstract

Diarrhoea, a disease that is extremely sensitive to water sources, is a leading cause of child mortality and morbidity and is widespread throughout developing countries. Meanwhile, climate change, along with extreme weather phenomena, is considered one of the main reasons causing water pollution. As a developing country located in the tropical climate belt, Vietnam has a high risk of diarrhoea outbreaks, which leads to an urgent need for an early warning system for this disease. However, previous studies have only focused on predicting diarrhoea incidence rates/cases using only a few traditional forecasting models with limited performances. In this paper, we focus on another problem of prediction diarrhoea outbreaks for six provinces in Vietnam. Our approach differs to all existing works as following. First, a wide range of 22 different machine learning models are thoroughly studied for the outbreak prediction task. This will provide a comprehensive picture on the overall performances of many different ML models on the outbreak prediction task for future research in other areas. Our study reveals that many of these models, e.g., SVM, XGBoost, and Decision Tree, can predict outbreaks more accurately than inferring them from predicted incidence rates/cases like previous researches. Moreover, diarrhoea outbreak prediction is a very challenging task where all existing algorithms are not capable of forecasting outbreaks well. Secondly, the main contribution of this paper is a proposed novel ensemble prediction method built upon the Apriori algorithm that could eliminate 10% to 40% of the mistakenly predicted outbreaks.