Dengue fever incidence forecasting : methodological approach in comparing SVR and LSTM models for design and implementation of a web-based forecasting framework

• Main Author: CO, ISABELLE-LYNN
• Format: text
• Published: Archīum Ateneo 2018
• Subjects: Dengue; Public health surveillance; System design; Data mining; Machine learning; Communicable diseases > Transmission > Prevention; Medical informatics.
• Online Access: https://archium.ateneo.edu/theses-dissertations/166; http://rizalls.lib.admu.edu.ph/#section=resource&resourceid=1535947331&currentIndex=0&view=fullDetailsDetailsTab
• Institution: Ateneo De Manila University

Dengue Fever is a mosquito-borne viral disease that has been a significant public health concern in the Philippines and many other tropical and subtropical regions in the world [42]. Among the numerous efforts in controlling Dengue Fever is strengthening of the countrys health surveillance systems presenting a more proactive approach in mitigating such cases. In this study, two data-driven machine learning models, namely Support Vector Regression (SVR) and Long Short Term Memory (LSTM), were developed without the requirement for prior epidemiological parameters and ordinary differential equations. Performances of the models in predicting Dengue incidences in the municipalities/cities of Western Visayas Region of the Philippines were mainly evaluated using Normalized Root Mean Square Error (NRMSE) and Pearson's r Correlation. Results showed that SVR performed generally better than LSTM and naive forecasts. From the evaluation cases performed, addition of feature selection shows plausibility of performance improvement especially on LSTM. Furthermore, experiments including ovitrap indices in the feature set showed evident performance increase. To realize its use for health surveillance, the resulting models were arranged in a web microservice accessible to the cloud-enabled system called Feasibility Analysis of Syndromic Surveillance Using a Spatio-Temporal Epidemiological ModeleR (FASSSTER) for Early Detection of Diseases.