Enhancing tourism demand forecasting with a transformer-based framework

This study introduces an innovative framework that harnesses the most recent 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. Our novel approach initially employs the decomposition method to decompose the data series to effectively mitigate the influence of outliers. The temporal fusion transformer is subsequently utilized for forecasting, and its hyperparameters are meticulously fine-tuned by a Bayesian-based algorithm, culminating in a more efficient and precise model for tourism demand forecasting. Our model surpasses existing state-of-the-art methodologies in terms of forecasting accuracy and robustness. [Display omitted] •A refined Transformer-based framework was introduced.•The accuracy of the standard transformer architecture was enhanced.•A robust decomposition approach was employed to improve forecasting.•The study also explores interpretability in deep learning.