Neural Network Forecasting: Error Magnitude and Directional Change Error Evaluation

摘要

In the case of tourism demand, better forecast would help directors and investors make operational, tactical, and strategic decisions. Time series forecasting methods can be divided into classical and modern methods. Classical methods here are the methods that already have establish procedures and do not involve artificial intelligence techniques. Most of thee classical methods are bounded with preliminary assumptions on the data distribution. On the other hand, modern methods do not require such assumptions. Neural networks (NN) can be considered as one of the most popular advance methods in time series forecasting as it has shown satisfactory performance in many literatures. In this study, the forecast performance of NN was compared with Box-Jenkins, time series regression, additive Holt-Winters and intervention model. Their forecast accuracies were evaluated by using error magnitude and directional change error measures. It was found that NN managed to forecast the data effectively in terms of both error measures and despite of the presence of outliers.