Prediction of dengue fever cases in Sri Lanka using time series model

Abstract

The number of dengue fever, one of the most dangerous mosquito viral infection, cases has dramatically increased in recent years. Since the first reported case of dengue fever in 1965, there had been occurrences on and off until the recent. Previous studies based on meteorological factors and the epidemiological pattern of cases are recommended for effective control programmes. Consequently, several programmes are conducted Island wide to control the cases. A mechanism is needed to estimate the number of cases and thus this study will provide support to forecast them. Therefore, the objective of the study is to develop a time series model to predict the dengue fever cases in future. Statistical tests are used to construct Auto Regressive Moving Averages (ARMA) models to predict the number of dengue cases. At preliminary stage, graphs of Auto Correlation Function, Partial Auto Correlation Function and Augmented Dickey Fuller test is used to test stationary of the series. For model selection, coefficient of determination, Durbin-Watson statistics, Akaike information criterion and Schwartz’s Bayesian criterion are used. Diagnostics tests on residuals are also carried out. Mean Absolute Percentage Error (MAPE) statistics is used to measure the accuracy of the model. The results reveal that 4 cases per day are recorded in Sri Lanka. Further, it shows that in two seasons: December- January and June- July; the cases are very high. The accuracy of the fitted model shows more than 75%. Therefore, the proposed model can be used to forecast the number of dengue cases in Sri Lanka.