Daily tourism demand forecasting via card transactions: a multi-source, interpretable, framework for diverse destinations and markets

Identificador:  imarina:9475710

Handle: https://hdl.handle.net/20.500.11797/imarina9475710

Autores:  Grau-Escolano, Jordi; Anton Clave, Salvador; Borras, Joan

Resumen:
This research introduces a novel comprehensive framework for accurate daily tourism demand forecasting by leveraging point-of-sale card transaction data and a variety of explanatory variables such as Google search trends, holiday data, and weather conditions. By employing a combination of statistical, machine learning, and deep learning models, the study generates daily forecasts across key urban and coastal localities in Catalonia, Spain, segmented according to tourists' nationalities. The results showcase the methodology's robust predictive power, maintaining high accuracy even amid the COVID-19 pandemic period, which was marked by severe uncertainties. Furthermore, the Temporal Fusion Transformer model's interpretability allows to provide valuable insights into the primary factors influencing predictions. The research further investigates the benefits of employing a unified forecasting model across various localities and nationalities, taking advantage of shared temporal patterns, finding that such an approach provides a more time-efficient training process without significantly compromising forecasting accuracy