Dengue fever outbreak prediction technique in area of Sisaket province using satellite-based climate and vegetation index

Abstract

This study aims to apply Seattleites based climate and vegetation index from NOAA polar-orbiting satellites to be predictor for build suitable dengue incidence rate predictive model of Sisaket province on the south area of northeastern of Thailand. The dataset used in this study is dengue severance report which contains records of patients found in Sisaket province area from January 2007 to December 2013 collected by the office of disease prevention and control, 10 Ubonrachathani. This data was preprocessed to compute incidence rate per 100,000 population. The Seattleites based climate and vegetation dataset is provided by NOAA polar-orbiting satellites. Both preprocessed dengue dataset and satellites dataset have been performing correlation analysis to find correlation and lag-time between dengue incident rate per week and each of climate and vegetation index. From correlation analysis we found SMN, SMT, VCI
และ
VHI with lag-time of 37, 21, 30 and 30 are suitable index and lag-time for use as predictor. After that 8 predictive models were built from those 4 index and lag-time with linear regression and support vector regression technique. Training sets were used to train predictive models and verify the most suitable model by calculating and comparing RMSE, MAPE and MAE of each model. The best predictive model for this study was built from SVR with 4 predictor SMN, SMT, VCI and VHI. Our predictive model can predict dengue incident rate for 21 weeks ahead. The proposed model verified predictive accuracy compared with traditional time series analysis techniques such as ARIMA.  The proposed predictive model significant better predict incident rate for 21 weeks ahead from week 32 to 52 in 2013 than ARIMA