Forecasting Performance Comparison of ARIMA and VAR Models in Dengue Cases Prediction for Five Districts in Sri Lanka

Abstract

Currently, there is an increase in the number of dengue cases reported in Sri Lanka. Therefore, the main objective of this study is to build a suitable time series model to predict the number of dengue cases in Sri Lanka. We considered the five districts, Colombo, Kurunegala, Nuwara Eliya, Trincomalee, and Puttalam, for the study. Also, we are interested in identifying a pattern of dengue cases in each district and studying the effect of climatic factors on dengue disease as the dengue vector mosquito lays its eggs in water. The data used in this study was the number of dengue cases reported in each district and the precipitation and temperature data as the climatic factors from 2010 January to 2022 June. Dengue data sets were collected from monthly reported dengue cases in Epidemiology Unit, Ministry of Health in Sri Lanka. Daily precipitation and temperature data were obtained from National Centers for Environmental Information which are publicly available, and monthly data was calculated. A comparative study of Vector Auto Regression (VAR) and Auto Regression Integrated Moving Average (ARIMA) in predicting dengue cases was carried out. The VAR and ARIMA models were fitted to the data predicted the dengue cases, and the accuracy by was checked by calculating Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percent Error (MAPE). Finally, the results of our study show that the VAR (1) model, which used only the climatic factor with a MAPE of 41%, is the best model to predict the dengue cases of Kurunegala. VAR (5) model with only the climatic factor precipitation with a MAPE of 51% is the best model to predict the dengue cases of Colombo, ARIMA (1,0,1) (1,0,0) [12] model with a MAPE of 40% is the best model for Nuwara Eliya. ARIMA (1,0,1) is the best model for the Puttalam and Trincomalee districts, with a 31% and 24% MAPE respectively.