A hybrid intelligent technique combines neural networks and tabu search methods for forecasting

Artificial neural networks (ANN) have been efficiently used in many fields. One of these fields is forecasting. Forecasting problem is an important issue on which many researchers from different disciplines have still working. ANN method has proved its success in time series forecasting. While traditional time series forecasting methods are not sufficient for some time series which have non linear structure, ANN method can produce very accurate forecasts for these time series. On the other hand, there are still some problems with using ANN in forecasting problems. ANN have some components such as architecture structure, learning algorithm and activation function which play important role on forecasting performance of ANN. Hence, determination of these components is a vital issue. Selection of architecture structure that consists of determining the numbers of neurons in the layers of a network is a key point for ANN method. To determine the best architecture which gives the most accurate forecasts, various approaches have been proposed in the literature. Finding the best architecture, which produce the best performance measure value, can be considered as an optimization problem. If the performance measure is a kind of criterion based on the error between the forecasts and the original values, this is a minimization problem. Aladag (2009) used tabu search algorithm to solve this problem so he proposed a hybrid method combines artificial neural networks and tabu search methods for forecasting problem. In this study, it is shown how the hybrid intelligent forecasting method in which tabu search algorithm is employed to determine the best architecture works. Then, the time series of international tourism demand of Turkey is forecasted by using the hybrid intelligent technique and obtained results are discussed.