Navigating Data Limitations: A Transfer Learning Compass for Accurate Tourism Demand Forecasting

Abstract

Accurate demand forecasting is essential for successful tourism management. Existing studies have improved forecasting performance focus on data preprocessing or model optimization, neglecting the challenge of efficient tourism demand forecasting with limited data. However, gathering sufficient data may be expensive or even impractical in real-world scenarios, especially for new or emerging destinations, or those lacking complete official statistics. To address the issue, this study introduces a forecasting method based on transfer learning and transformer neural network. Additionally, the approach incorporates a two-stage data selection strategy and a model ensemble technique to further enhance forecasting performance. Experimental results suggest that the proposed method generates accurate forecasts with limited data availability, and significantly outperforms popular benchmarks. Moreover, our method shows excellent robustness and scalability.