A 2-layered game-theoretic deep learning model for predicting lodging demand: SHAP and nash equilibrium

Abstract Forecasting tourism demand is critical for pricing and revenue management, yet most advanced machine learning models operate as “black boxes,” limiting interpretability and trust. This study introduces an explainable, game-theoretic hybrid deep learning framework for predicting short-term rental demand. This study incorporates game theory in two layers: a cooperative game and a competitive game. Using 729,888 daily observations from 1,852 Airbnb listings in Ann Arbor, Michigan, we compare traditional models (regression, SARIMAX) with advanced architectures (FCNN, LSTM, Transformer, and a Transformer–LSTM hybrid). Game theory is used in two layers: cooperative game (SHAP) and competitive game (classic Nash equilibrium). SHAP creates an explainable deep learning model to quantify each feature’s contribution. The best-performing model (FCNN) was further used to simulate a 3x3 pricing game between pairs within 4 submarkets. Results show that deep learning models significantly outperform traditional baselines, while SHAP-based feature attribution and clustering enhance transparency and strategic insight. The competitive games shed light on the characteristics and dynamics of each submarket. This research offers insights to revenue managers, policymakers, and operators.