Predicting Airbnb occupancy rate: a hybrid deep learning and game-theoretic approach

Purpose In this study, the authors used advanced machine learning (ML) models combined with game theory concepts to predict the occupancy of Airbnb properties. By leveraging big data and explainable artificial intelligence (AI), this study aims to identify the most accurate and interpretable model for occupancy prediction. Design/methodology/approach This study applies advanced ML models to predict occupancy rates using a comprehensive data set containing 729,888 data points from 1,852 distinct properties in Ann Arbor, Michigan. By examining key features such as property features, amenities, external events and spatial characteristics, the authors assess the performance of models through metrics including root mean squared error (RMSE), MAE, mean squared error (MSE) and mean absolute percentage error (MAPE). Findings The authors showed that a game-theoretic hybrid model (long- and short-term memory + transformer) outperformed traditional and base ML models. The integration of the game-theoretic concept of Shapley values enhances the model transparency by quantifying the impact of each predictor, particularly external factors and spatial attributes. Research limitations/implications This study advances game theory and deep learning applications in the hospitality and tourism forecasting literature. Originality/value To the best of the authors’ knowledge, this study is the first in hospitality research to integrate big data, deep learning, game theory and explainable AI for occupancy prediction. The novel hybrid model proposed in this study overcomes the shortcomings of the standalone model architectures, such as a lack of transparency and sequential dependency. Hence, the proposed model captures both long-term trends and short-term fluctuations of data.