Dengue Incidence Prediction Using Model Variables with Registered Case Feedback

Abstract

This study discussed building of localized dengue incidence prediction models for districts of Selangor. System identification with Linear Least Square estimation method is used to build a number of model orders with varied lag-time and the most accurate model is selected for each district. Model accuracy is measured using Mean Square Error (MSE) value, with smaller MSE value, represents better accuracy. The flow of study is started with identification of significant weather variables. It was found that all three weather variables namely mean temperature, relative humidity and rainfall are significant predictors. Further inclusion of dengue incidences feedback data into the model was found to enhance the model accuracy. Model accuracy is further tested by comparing between single and ensemble model of few districts. Ensemble model is built using dengue prediction model of its district together with its neighbouring districts, and was found to be better predictor in two out three districts. Therefore, it was concluded that ensemble models predict better in some cases, and single models are better in other cases, depending on rate of human movement between neighbouring districts.