CLARION-RF: A Climate-Lag-Aware Probabilistic Framework for Early Prediction of Vector-Borne Disease Outbreak Risk in Karnataka, India

Abstract

Vector-borne diseases like dengue, chikungunya, and malaria continue to be a significant public health concern in India, especially in climatically sensitive areas where seasonal effects have a strong impact on disease transmission. Existing national surveillance systems are largely reactive, detecting outbreaks only after case counts rise substantially. This work proposes the CLARION-RF model, a probabilistic framework that is climate-lag aware and capable of early outbreak-risk prediction at the district level. The model combines time-series data from the weekly disease surveillance system with corresponding meteorological variables and climate lag features. The Random Forest-based probabilistic model is trained by balancing the samples to address class imbalance. Instead of deterministic predictions, the model estimates outbreak probabilities and stratifies them into three risk levels: low, medium, and high. Time-aware training (2011–2023) and testing in 2024 demonstrate stable predictive performance, with ROC-AUC and PR-AUC of 0.667 and 0.579, respectively, for the best model.