| Global health security depends on the early identification of biological hazards, but conventional surveillance systems often rely on delayed indicators such as reported cases and hospitalizations. This work presents a multimodal learning framework that integrates genomic surveillance, epidemiological reports, and population mobility trends for early biological threat detection. On retrospective U.S. COVID-19 surveillance data with 2.6% threat prevalence, the proposed model trained on 6,305 integrated samples achieved a precision of 0.94, recall of 0.96, and F1-score of 0.95, outperforming selected baseline models. SHAP analysis identified influential predictors, including variant-related features and anomalous mobility reductions. The framework demonstrated consistent performance across temporal validation folds over 49 weeks of data. The evaluation is limited to retrospective COVID-19 data from the United States, and further validation is required for other diseases and regions. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.