Dengue outbreak prediction using geographic information system and machine learning: A case study of Mueang Phayao District, Phayao Province

Abstract

Dengue fever is a mosquito-borne infectious disease and is a public health problem in many countries around the world. This research aims to study the factors affecting the spread of dengue fever and to create prediction models using machine learning techniques. The data were collected from the Phayao Meteorological Station and Phayao Hospital between 2018 and 2022. The geographic information system was used to analyze the related factors. The data preparation was performed by separating the datasets according to various factors,

including gender, age group, population, forest area, distance between outbreak areas and water sources, distance between re-outbreak areas, average temperature, average relative humidity, average rainfall, and number of rainy days.

                All data were generated as monthly and weekly datasets. The prediction models are divided into regression and classification types. The regression includes support vector machines, generalized linear models, and artificial neural networks. The classification includes decision trees, gradient boosting trees, and random forests. The ensemble method of stacking models is used to combine multiple machine learning algorithms to improve the performance by using 2 algorithms as base learners and using another algorithm as the stacking model learner. Performance evaluation was performed through five-fold cross-validation and the different datasets and learning methods were compared.

                The results found that the regression type using the generalized linear model as the stacking model learner gave the best performance with an RMSE value of 1.457, while the classification using the gradient boosting tree gave the highest performance with an accuracy of 99.12%. This study highlights the importance of using machine learning techniques in forecasting dengue fever outbreaks, which can be used as an effective tool for disease surveillance and control in the future.