Comparative Analysis of Machine Learning and Deep Learning Models for Tourism Demand Forecasting with Economic Indicators

This study addresses the critical need for accurate tourism demand forecasting in Bul-garia using economic indicators, particularly following COVID-19's demonstration of the sector's vulnerability to systemic disruptions. The research employs ensemble ma-chine and deep learning methodologies, combining Prophet with external regressors, Ridge regression, and gradient boosting models using inverse MAE weighting optimi-zation. Using monthly overnight stay data from Bulgaria's National Statistical Institute (2005-2024) integrated with COVID-19 cases and Consumer Price Index (CPI) indica-tors, the study reveals varying ensemble performance across different implementa-tions. Initial evaluation showed the ensemble model achieving MAE of 156,847, RMSE of 298,245, and MAPE of 14.23%, outperforming individual models by 10.2%. Howev-er, comprehensive testing revealed different characteristics: the Feedforward + Proph-et Ensemble performed best with MAE of 762,868 and MAPE of 58.02%, while tradi-tional Prophet (Seasonal Only) showed MAE of 910,000 and MAPE of 72.80%. Com-plex architectures like BiLSTM + MultiHead Attention achieved MAE of 875,129 but exhibited negative R² scores, suggesting overfitting. Performance variation across evaluations highlights dataset dependency and con-text-specific model selection importance. The ensemble approach consistently main-tained competitive performance, providing enhanced forecasting capability for tour-ism stakeholders' investment planning, marketing budgets, and operational capacity decisions. Economic indicator integration effectively captures structural breaks in tourism patterns, offering practical insights for robust demand forecasting during economic volatility.