Generation of ensemble forecasts using functional-link net for decomposition ensemble learning to forecast tourist arrivals

Decomposition and ensemble approaches have come to play an increasingly significant role in the tourism literature. Nonlinear artificial intelligence (AI) methods, including the multi-layer perceptron and support vector regression, have been widely employed to predict the demand for tourism based on decomposition and ensemble learning. However, even though the effectiveness of functional-link net (FLN) with a flat network structure at function approximation, few of studies are devoted to the use of FLN to tourist arrivals forecasting. The main issue we address is to investigate the effect of combining different decomposition methods with FLN on the forecasting performance of decomposition ensemble models. This study features the development of decomposition ensemble models to forecast the tourist arrivals by using commonly used AI methods to forecast the series of individual modes after decomposing data into several subsequences, and FLN was then applied to generate ensemble forecasts. We used the inbound tourist arrivals to Taiwan to verify the performance of the proposed decomposition ensemble methods. It was found that they significantly outperformed commonly used benchmark models.