This study examines how to improve influenza forecasting by combining mechanistic epidemiological models with machine-learning ensemble methods. The researchers developed a “random forest of epidemiological models” framework where outputs of multiple mechanistic models (which simulate disease dynamics) serve as input features into a tree-ensemble (random forest) algorithm that learns to predict future influenza incidence more accurately. They applied the framework to historical influenza data, comparing its forecasting performance against the mechanistic models alone and against standard machine-learning methods that do not incorporate mechanistic model outputs. The findings revealed that the ensemble approach achieved significantly better forecast accuracy (lower error metrics) and greater consistency across different years and regions. In their discussion, the authors highlight that leveraging mechanistic model insights within a machine-learning ensemble bridges the gap between theory-driven and data-driven methods, and they suggest that such hybrid modelling frameworks could be generalized to other infectious disease contexts.