AI-Driven Tourism Demand Forecasting Using Multi-Source Mobility, Crisis, and Sentiment Data

This study proposes a crisis-aware, AI-driven framework for short-term tourism demand forecasting using multi-source digital traces in India, the United States, and France. It fuses mobility indicators, crisis-event intensity from GDELT, online sentiment from TripAdvisor reviews, and metadata on accommodation supply from Airbnb to embed behavioral, perceptual, and disruption-related drivers of tourism activity within the COVID-19 pandemic recovery period. A hybrid forecasting pipeline based on Prophet with interpretable regressors was developed and compared to a PCAbased model and an ensemble approach. Findings indicated that the parsimonious regressor model considering winzorization based on crisis events yielded the best performance, returning the lowest MAE and RMSE. PCA components decreased responsiveness in case of abrupt demand surges. The main finding is that multimodal, crisis-sensitive features are highly relevant to improving forecast accuracy under turbulent conditions. This framework offers a practical basis for developing a real-time, crisis-aware tourism-forecasting system to support data-driven decisions in highly uncertain environments.