River Level Forecasting

Description:

This project proposes an ensemble of Recurrent Neural Network models to forecast river flow rate using weather data from throughout the catchment. Small, localized models are trained for each weather data source and predictions are combined to predict the flow rate at a chosen site. This ensemble approach allows for a larger input dataset while avoiding the pitfalls of large dimensionality. The final ensemble model yields a MAPE of 8.57%, which is 10.02% lower than the best performing localized model.