AI-Driven Smart Tourism: Leveraging Deep Learning and Financial Econometrics for Predictive Risk Management and Dynamic Destination Optimization

Background: The tourism industry faces unprecedented volatility from economic fluctuations, geopolitical tensions, and health crises, necessitating advanced analytical approaches beyond traditional econometric methods. While ARIMA, GARCH models, and extreme value theory provide foundational frameworks, integrating deep learning with financial econometrics presents transformative opportunities for smart tourism destinations (STDs) to enhance predictive risk management and dynamic optimization capabilities. Methods: Following PRISMA guidelines, we systematically reviewed studies published between January 2020 and 2025 across six academic databases. From 4327 initial records, 63 high-quality studies met inclusion criteria. Analysis focused on four research questions: (1) integration of deep learning with financial econometric models (VaR, CVaR, GARCH); (2) multi-source data fusion strategies; (3) portfolio theory applications; and (4) real-time cascading risk prediction. Results: Hybrid architectures combining deep learning with financial risk models achieved 20–45% performance improvements over traditional approaches. SAE-Bi-GRU and SARIMA-CNN-LSTM models demonstrated superior accuracy. Internet search data emerged as the most powerful predictor, while multi-source fusion enhanced forecasting accuracy by 30–45%. However, only 5 of 33 deep learning studies explicitly addressed risk management. Portfolio-theoretic approaches remain unexplored, and real-time adaptive frameworks are absent. COVID-19 catalyzed innovation, with AI-driven destinations recovering 40–60% faster. Extreme value theory applications achieved 87% accuracy in predicting extreme tourism events. Conclusions: Integrating deep learning with financial econometrics offers transformative potential for resilient STDs. Future research should develop integrated risk-aware frameworks combining stochastic models (LSTM, GARCH) with portfolio optimization, implement federated learning for privacy-preserving collaboration, and establish multi-objective optimization balancing economic returns with sustainability. Standardized evaluation protocols are needed to advance the field toward intelligent tourism management systems capable of thriving amid uncertainty.