Cross-border tourism in North America: A hybrid deep learning framework with macroeconomic indicators

This study proposes a hybrid predictive framework designed to forecast border tourism flows among the United States, Canada, and Mexico. Combining Fuzzy Markov Chains, Hidden Markov Models, and attention-based deep learning architectures (RNNs, GRUs, and CNNs), the model captures the complex and dynamic interplay between exchange rate volatility and broader macroeconomic conditions over forty years. Results show that tourist behavior is shaped by both current economic indicators and long-term economic memory, with attention mechanisms offering interpretable insights into spending and arrival trends. The SUOS model outperforms traditional forecasting approaches, demonstrating superior accuracy and scalability. Its interpretability also enables stakeholders to understand which economic factors drive tourism demand, supporting practical applications such as seasonal planning, marketing timing, and policy formulation. By bridging macroeconomic modeling with advanced AI, this research offers a robust and adaptive tool for anticipating tourism shifts in an increasingly uncertain global economy.