Forecasting Dengue Outbreaks in Kalutara District using Hybrid Machine Learning Techniques

Dengue is the most rapidly expanding mosquito–borne viral disease that has emerged as a significant global public health concern in tropical and subtropical regions. Almost all districts in Sri Lanka, one of the dominant countries affected by the dengue epidemic, are announcing dengue cases, and Colombo, Gampaha and Kalutara districts in the western province have recorded the highest case rate. Ongoing prevention efforts aim to reduce dengue risk, and accurate forecasting is vital for effective surveillance and control. This study demonstrates that a hybrid machine learning approach improves vector-borne disease forecasting accuracy. The spread of dengue depends on factors such as weather, urbanization, population density, water storage containers and tanks, etc. This study aims to design an Artificial Neural Network (ANN) model utilizing the Backpropagation (BP) algorithm and Levenberg-Marquardt (LM) algorithm independently, to forecast the average number of Dengue cases for next week in Kalutara District where one of the highest dengue-reported districts in Sri Lanka, based on various weather factors, and enhance the accuracy of the models by applying gradient boosting method. The model's input and output were selected as the current week's climate factors and the number of Dengue cases for the following week in the Kalutara district, respectively. Weekly data from 2022 January to 2023 June, a period with minimal external impact due to COVID-19, were collected from District Secretariat Kalutara, and NASA POWER Project which provides solar and meteorological data sets from NASA research. Parameters and accuracy of the models were measured based on minimum values of mean square error (MSE), root mean square error (RMSE) and mean absolute error (MAE). While the ANN model with the LM algorithm initially performed well (MSE=0.012166, RMSE=0.113006, and MAE=0.069093), applying gradient boosting improved model accuracies, and the BP algorithm later achieved high accuracy (MSE=0.000613, RMSE=0.025006 and MAE=0.017425). These results show that hybrid machine learning algorithms effectively enhance the accuracy of vector-borne disease forecasting.