Abstract
Dengue remains a significant and critical global health concern, especially in resource-constrained and remote regions, where traditional IgG/IgM-based testing is often delayed or not conducted properly. Furthermore, conventional machine learning often exhibits minimal interpretability and misclassification, leading to major unreliability in real-time clinical decisions. To tackle these hindrances, we proposed an interpretable, efficient, and novel machine learning framework that operates near real-time. It combines feature optimization using Genetic Algorithms (GA) and Generative Adversarial Networks (GAN) to address data imbalance, and enhances ubiquitous decision interpretability with Explainable AI (XAI). GA establishes the most predictive hematological features, which improve accuracy and transparency, whereas GAN-based data generation handles class imbalance, leading to enhanced generalization. On top of that, the optimized Decision Tree model attains 99.49% accuracy with a negligible computational cost of training and testing time 0.0025 s, and 0.0013 s respectively, superseding the current state-of-the-art. A web-based application implemented based on the proposed model enables real-time risk prediction with a latency of under 0.6 s. A comprehensive XAI evaluation using LIME, SHAP, Morris sensitivity analysis, permutation combination, and RFE consistently identifies WBC and platelet counts as key predictors. In numbers, XAI techniques represent that low White Blood Cell (WBC) count (< 3,700 cells/μL), platelet count (< 136,000 cells/μL), and Platelet Distribution Width (PDW < 23) are key indicators of dengue. Our proposed integrated GA-GAN-XAI framework bridges accuracy, interpretability, and real-time decision-making capability. This approach is highly accurate, robust for healthcare, and a highly deployable solution for dengue risk prediction for clinical dengue risk assessment.