Khaled Shaalan

This paper describes a neural network approach for time series data forecasting. This study looks at the application of neural computing to an agricultural problem. In the selected application, the first step was to construct the multilayer feedforward neural network and the backpropagation training was employed. Elements of preprocessing, preparatory work, and network design are discussed. In order to improve efficiency and achieve stability, experiments were done while varying the learning rate and the number of hidden units. As a result, successful generalization has been observed on the test sets. Also a statistical approach (regression analysis) was applied to the same application. A comparative study between the applied methods was described. The comparative study can also be beneficial for implementing other forecasting applications. Our study suggests that the neural network is very useful for forecasting problems and in terms of comparative accuracy, the network was able to achieve more accurate results than the multipe regression model.