Neural networks in forecasting models: Nile River application

The neural network approach is applied to the prediction of the flow of the River Nile. A multilayer feedforward network is constructed and trained by the backpropagation algorithm. We propose several different methods for single-step ahead forecast and multi-step ahead forecast in an attempt to get the least prediction error. These methods investigate different ways to preprocess the inputs and the outputs. We consider ten-days ahead forecast and one-month ahead forecast. In both cases good results were observed.