Modeling a combined forecast algorithm based on sequence patterns and near characteristics: An application for tourism demand forecasting

• Published in: Chaos, solitons and fractals Vol. 108; pp. 136 - 147
• Main Authors: Jun, Wang, Yuyan, Luo, Lingyu, Tang, Peng, Ge
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.03.2018
• Subjects: Artificial neural network; Combined forecasting method; K-means clustering; Tourism demand forecasting; Artificial neural network; Combined forecasting method; Tourism demand forecasting; K-means clustering
• ISSN: 0960-0779; 1873-2887
• DOI: 10.1016/j.chaos.2018.01.028

Tourism demand forecasting is essential for forward tourism planning. To develop appropriate public policies and ensure sound business investment decisions, both government administrations and private sector businesses use basic tourist demand forecasting to plan future operations and assess the need for facilities and infrastructure investment. Therefore, forecasting has become indispensable to tourism management. This study proposes a combined tourism forecasting model using an artificial neural network (ANN) and a clustering algorithm, which considers two aspects of the given data series: sequence patterns and near characteristics, which embody structural changes and time series correlations. Training data were clustered into homogenous groups, and for each cluster, a dedicated forecaster was employed. Several neighboring samples were then selected to capture the current changes in the data series trends. Finally, the two prediction results derived from the sequence patterns and near characteristics were combined to determine the final forecast results. To verify the superiority and accuracy of the proposed model, it was compared with three other ANN-based models and the most popular ARIMA model using three non-linear, non-stationary tourist arrivals data series. Experimental cases studies demonstrated that the proposed combination method consistently outperformed the other related methods.