Tourism demand forecasting with multi-source data: a hybrid model incorporating seasonal-trend decomposition

Zhen Zhang, Ziyu Xu, Wenyu Yu

Abstract: Accurate daily forecasting of tourism demand is crucial for enhancing the efficiency of tourist attraction management, improving tourist experiences, and promoting sustainable development. This paper introduces a tourism demand forecasting model that integrates seasonal-trend decomposition with multi-source data, including search engine data, weather data, and holiday data. The proposed model comprises three key steps: (1) using Seasonal-Trend decomposition based on Loess smoothing to decompose the historical tourist arrival sequence; (2) analyzing the correlation within the search engine data and applying principal component analysis to reduce its dimensionality; and (3) employing a long-short term memory network to model each sub-time series derived from the decomposition based on multi-source data, while optimizing model parameters through Bayesian optimization. To evaluate the effectiveness of our model, we apply it to predict daily tourist arrivals at Jiuzhaigou and Mount Siguniang across various prediction horizons. The experimental results demonstrate that our model significantly improves the accuracy of tourism demand forecasting.

External IDs:doi:10.1007/s10479-025-06832-0