Assessing machine learning methods in predicting dengue incidence using climatic factors in Region IV-A (CALABARZON), Philippines

• Main Authors: Castro, Ian Kevin G., Elquiero, Nikki Elisha M., Fradejas, Jericho D.
• Format: text
• Language: English
• Published: Animo Repository 2023
• Subjects: Dengue > Philippines; Machine learning; Biology
• Online Access: https://animorepository.dlsu.edu.ph/etdb_bio/27; https://animorepository.dlsu.edu.ph/context/etdb_bio/article/1028/viewcontent/2023_Castro_Elquiero_Fradejas_Assessing_machine_learning_methods_in_predicting_dengue_incidence_Full_text.pdf
• Institution: De La Salle University

Region IV-A or CALABARZON in the Philippines records a high number of dengue cases annually. Several studies worldwide have used machine learning techniques using climatic factors to forecast dengue outbreaks. In this study, the performance of six machine learning models namely (a) Random Forest, (b) Conditional Inference Forest, (c) Extreme Gradient Boosting, (d) Support Vector Machines, (e) Least Absolute Shrinkage and Selection Operator, and (f) Generalized Additive Modeling were compared by predicting dengue incidences associated with climatic factors (temperature, precipitation, and relative humidity). The datasets, both with and without delayed effects, were subjected to different modeling techniques and evaluated based on the root mean square error, mean absolute error, and correlation coefficients. Relative humidity was the climatic factor that had the highest correlation with dengue incidences while temperature was the factor with the weakest correlation. The models with delayed effect generated the highest predictive accuracy in all machine learning methods except for Random Forest, Conditional Inference Forest, and Extreme Gradient Boosting in the spatial scale with all barangays of CALABARZON. Furthermore, Random Forest with delayed effect was deemed as the best model for all spatial scales except for all barangays, which had Random Forest without delayed effect as the best model. The study demonstrated the potential of using machine learning methods in forecasting dengue outbreaks and creating dengue prevention programs in Region IV-A yet there is a great need for more studies within the region.