An enhanced lstm framework for arboviral outbreak prediction using multisource data fusion and multitask temporal learning

Arboviral infections, including dengue, Zika, and chikungunya, pose persistent public-health challenges because transmission depends on interacting epidemiological, climatic, vector-related, environmental, and population factors. This study presents a multisource, multitask long short-term memory (LSTM) framework intended to produce two disease-specific outputs: binary outbreak alerts and forecasts of case counts. The proposed design aligns heterogeneous predictors into multivariate temporal sequences and uses shared temporal representations with separate prediction heads. The supplied results table reports mean classification accuracy of 92.73%, mean F1-score of 91.22%, mean absolute error of 10.07 cases, and root mean squared error of 14.13 cases across the three diseases. A separate comparison table reports a lead time of approximately 14 days. These figures are presented as previously reported results, not independently reproduced values: the underlying dataset identifiers, final test predictions, outbreak labels, model configuration, and lead-time event records must be verified before submission. The practical value of the approach rests on whether such performance generalizes to genuinely prospective surveillance settings under realistic reporting delays and uneven data availability.