Neural networks for web server workload forecasting

This paper presents a comparative study of five intelligent forecast models for workload of server defined as HTTP requests. These five forecast models are based on the methodology: Nonlinear AutoRegressive model with eXogenous Inputs (NARX), Multilayer Perceptron (MLP), Elman, Cascade-Neural Network (CCNN) and Pattern Recognition Neural Network (PRNN). The best accuracy prediction is given by the NARX model. This work takes parts in development of our forecast models in the project EnergeTic-FUI, France.