This study introduces an innovative framework that leverages the latest transformer architecture to enhance tourism demand forecasting. The proposed transformer-based model integrates the tree-structured parzen estimator for hyperparameter optimization, a robust time series decomposition approach, and a temporal fusion transformer for multivariate time series prediction. The model first decomposes the data series to mitigate the influence of outliers, then uses the temporal fusion transformer for forecasting, with hyperparameters fine-tuned via Bayesian optimization. The model outperforms existing state-of-the-art methods in forecasting accuracy and robustness. Tourism demand forecasting is crucial for strategic planning, resource allocation, and decision-making in the tourism sector. Traditional time series and econometric methods face challenges in capturing nonlinear patterns, seasonality, and outliers. Recent advances in artificial intelligence, including deep learning models like recurrent neural networks and long short-term memory networks, have shown promise but face issues like overfitting and lack of interpretability. The temporal fusion transformer, a state-of-the-art deep learning model, excels in capturing complex temporal dependencies and seasonal patterns but requires careful hyperparameter tuning. To address these challenges, this study proposes a novel transformer-based model that combines tree-structured parzen estimator for hyperparameter optimization, robust time series decomposition, and temporal fusion transformer for multivariate time series prediction. The model decomposes the time series into trend, seasonal, and residual components, then uses the temporal fusion transformer for forecasting with Bayesian optimization. The model is evaluated on three datasets and shows superior performance in forecasting accuracy and robustness. The study also highlights the importance of interpretability in artificial intelligence models, particularly in tourism demand forecasting where decision-makers need transparent models. The proposed model's attention mechanism and variable importance analysis provide insights into the factors influencing tourism demand, enhancing the model's interpretability and usefulness for decision-making. The results demonstrate that the proposed model significantly improves forecasting accuracy and robustness compared to existing methods, offering a valuable solution for tourism demand forecasting.This study introduces an innovative framework that leverages the latest transformer architecture to enhance tourism demand forecasting. The proposed transformer-based model integrates the tree-structured parzen estimator for hyperparameter optimization, a robust time series decomposition approach, and a temporal fusion transformer for multivariate time series prediction. The model first decomposes the data series to mitigate the influence of outliers, then uses the temporal fusion transformer for forecasting, with hyperparameters fine-tuned via Bayesian optimization. The model outperforms existing state-of-the-art methods in forecasting accuracy and robustness. Tourism demand forecasting is crucial for strategic planning, resource allocation, and decision-making in the tourism sector. Traditional time series and econometric methods face challenges in capturing nonlinear patterns, seasonality, and outliers. Recent advances in artificial intelligence, including deep learning models like recurrent neural networks and long short-term memory networks, have shown promise but face issues like overfitting and lack of interpretability. The temporal fusion transformer, a state-of-the-art deep learning model, excels in capturing complex temporal dependencies and seasonal patterns but requires careful hyperparameter tuning. To address these challenges, this study proposes a novel transformer-based model that combines tree-structured parzen estimator for hyperparameter optimization, robust time series decomposition, and temporal fusion transformer for multivariate time series prediction. The model decomposes the time series into trend, seasonal, and residual components, then uses the temporal fusion transformer for forecasting with Bayesian optimization. The model is evaluated on three datasets and shows superior performance in forecasting accuracy and robustness. The study also highlights the importance of interpretability in artificial intelligence models, particularly in tourism demand forecasting where decision-makers need transparent models. The proposed model's attention mechanism and variable importance analysis provide insights into the factors influencing tourism demand, enhancing the model's interpretability and usefulness for decision-making. The results demonstrate that the proposed model significantly improves forecasting accuracy and robustness compared to existing methods, offering a valuable solution for tourism demand forecasting.