Interpretable tourism demand forecasting with temporal fusion transformers amid COVID-19

Area of Science:

• * Artificial Intelligence
• * Data Science
• * Tourism Management

Background:

• * Existing tourism demand forecasting models often lack interpretability.
• * The COVID-19 pandemic significantly impacted global travel patterns and necessitated more accurate forecasting methods.

Purpose of the Study:

• * To develop an interpretable tourism demand forecasting model.
• * To enhance prediction accuracy by incorporating diverse data sources, including sentiment analysis.
• * To optimize the Temporal Fusion Transformer (TFT) model using the Adaptive Differential Evolution (ADE) algorithm.

Main Methods:

• * Development of an Adaptive Differential Evolution-Temporal Fusion Transformer (ADE-TFT) model.
• * Utilizing historical tourism data, COVID-19 case numbers, and big data from travel forums/search engines.
• * Employing a Convolutional Neural Network (CNN) for traveler mood analysis and Latent Dirichlet Allocation (LDA) for topic modeling of travel-related discussions.
• * Keyword extraction from Google Trends data.

Main Results:

• * The ADE-TFT model demonstrated improved interpretability in tourism demand forecasting.
• * Integration of quantitative data and traveler sentiment analysis enhanced forecasting precision during the pandemic.
• * Attention analysis and input factor ranking provided insights into prediction drivers.

Conclusions:

• * The proposed ADE-TFT model offers a significant advancement in interpretable tourism demand forecasting.
• * Combining quantitative metrics with qualitative, sentiment-based features is effective for predicting tourism demand amidst crises.
• * The methodology provides valuable tools for understanding and forecasting travel behavior in dynamic environments.