Prediction of Transmission Rates of Dengue in National Capital Territory Delhi Using Machine Learning Models

The use of Machine Learning (ML) algorithms for predictive modeling to monitor transmission rates of dengue has gained significant attention worldwide. Earlier research has focused on specific weather variables and algorithms, there is a significant demand for models incorporating a wide range of variables and algorithms for superior performance. This study aims to predict transmission rates of dengue at the ward level in National Capital Territory (NCT) Delhi, India using different ML models. Data regarding incidence of dengue along with population and meteorological data as predictors during the period 2015-2022 have been used. The incidence data of dengue and population data have been collected from the Municipal Corporation Delhi (MCD) and Census of India, respectively. Meteorological data consisting of 20 parameters has been downloaded from NASA POWER. Five ML algorithms, including an ensemble approach, have been trained and validated. Comparative assessments using the Receiver Operating Characteristic (ROC) with the Area Under the Curve (AUC), accuracy, and F1 score have been carried out. It has been found that the accuracy of ensemble ML methods, such as Gradient Boosting, and Random Forest outperformed other models, such as Logistic Regression, Decision Tree, and Support Vector Machine. A correlation coefficient (r) of 0.40 has been used to evaluate the influence of meteorological variables on dengue transmission. Variables that exceed this threshold are considered to make a significant contribution to the transmission of dengue. Variable importance analysis shows significant contribution of Surface Soil Wetness (r= 0.48), Relative Humidity at 2 Meters (r= 0.45), Precipitation (r= 0.42), Wind Direction at 2 Meters (r= -0.54), Wind Direction at 10 Meters (r= -0.53), Temperature at 2 Meters (r= -0.47), and Earth Skin Temperature (r= -0.46). This study provides an excellent basis for future research, notably on dengue transmission modeling in dense urban environments and early warning systems using predictive models. This study provides insight on how ML algorithms may forecast dengue transmission rates and emphasizes the need to examine a variety of variables for model success.