A1057
Title: Probabilistic deep learning for forecasting influenza hospitalizations
Authors: Itai Dattner - University of Haifa (Israel) [presenting]
David Shulman - University of Haifa (Israel)
Rami Yaari - Columbia University (United States)
Jeffrey Shaman - Columbia University (United States)
Abstract: Accurate forecasting of hospitalizations due to influenza is essential for effective public health planning and response. In the U.S., the CDCs FluSight Forecast Hub collects and compares real-time forecasts of influenza-associated hospitalizations from various modeling teams. However, many models face trade-offs between predictive accuracy and interpretability. A deep learning framework is proposed that generates probabilistic forecasts of influenza-related hospitalizations across U.S. states by parameterizing a negative binomial distribution through a neural network. The approach captures both spatial and temporal variability while providing full predictive distributions. Model explainability and theoretical guarantees are discussed for the proposed methodology. Preliminary results indicate that the approach achieves state-of-the-art forecasting accuracy, supporting data-driven public health decision-making.
