Improved multi-step prediction of daily tourism demand: an innovative hybrid machine learning framework with search engine data

Tourism demand forecasting is a vital component of strategic planning in the tourism industry. This study proposes an innovative hybrid framework that integrates variational mode decomposition with a modified sparrow search algorithm and kernel extreme learning machine. The hybrid model incorporates weather factors and search engine data, including both Google Trends and YouTube Index, to enrich its predictive capacity. This research contributes to multistep tourism demand prediction, enabling daily, weekly, biweekly, and monthly scheduling planning. Using tourist arrival data from Hawaii and New Zealand as case studies, the single- and multi-step prediction results reveal that data decomposition and parameter optimisation significantly improve predictive accuracy. Furthermore, we incorporate explainable artificial intelligence to quantify the importance of influencing factors, enhancing the transparency and practical relevance of the model. To further assess robustness, we analyse forecasting performance across different periods of the COVID-19 pandemic. The findings provide valuable insights for improving management efficiency, adapting to dynamic environments, and leveraging search engine data effectively.