A high-resolution GIS and machine learning approach for targeted disease management and localized risk assessment in an urban setup: A case study from Bhopal, Central India

Predicting dengue distribution based on environmental factors is crucial for effective vector control and management as environmental factors like temperature, demographics, and artificial changes such as roads and buildings significantly influence dengue distribution. The use of new, emerging machine-learning techniques can aid in accurately predicting these cases and developing early warning systems. In this study, we divided our study area, Bhopal city, into 643 polygons of one square kilometre area and collected data on environmental and other factors. Dengue cases from 2012 to 2022 were mapped into these units and divided them into five categories. To find the best predictive model, we evaluated popular machine learning algorithms such as support vector machine (SVM), logistic regression, neural networks, random forest, k-Nearest Neighbors (kNN), and tree using parameters like area under the receiver operating characteristic (ROC) curve (AUC), classification accuracy (CA), F1 score, precision, and recall. The neural network performed the best, with an AUC of 0.921, CA of 0.755, F1 score of 0.740, precision of 0.732, and recall value of 0.755 and was thus selected for future predictions. Among the predictors, building area, population and road density had the highest influence, followed by minimum, maximum, and average temperatures in decreasing order of importance. The machine learning approach neural network effectively predicted the historical dengue distribution considering both landscape and climatic variables for an urban settings like Bhopal. This approach holds potential for application in other cities as well, highlighting the increasing importance of machine learning and predictive modelling in public health.