Bridging the predictive divide: A hybrid early warning system for scalable and real-time dengue surveillance in LMICs.

The global resurgence of dengue presents an ongoing challenge for public health systems, particularly in low- and middle-income countries (LMICs) where conventional early warning systems (EWS) often suffer from reporting delays and under-detection. While AI-powered EWS offer superior accuracy, their reliance on dense data streams and advanced infrastructure limits their scalability in resource-limited contexts. This paper introduces a hybrid EWS architecture that strategically combines retrospective epidemiological data with selective real-time signals-such as climate variables and digital trends-within a modular machine learning framework. Drawing on case studies from Brazil, Malaysia, and Vietnam, we demonstrate how this architecture adapts to diverse data environments: integrating seroprevalence data to correct underreporting, enhancing zoning-based alerts with behavioral signals, and using climate predictors to overcome data fragmentation. Simulation results indicate that the hybrid model reduces outbreak response time from 7 to 14 days (traditional EWS) to 3-5 days and improves prediction accuracy to 85-90 %. These findings highlight the hybrid EWS as a context-sensitive, scalable solution that balances predictive performance with implementation feasibility-offering a viable pathway for LMICs to operationalize real-time dengue surveillance and proactive vector control.