Spatial fairness in taxi demand prediction: a hybrid approach leveraging district profiles

Accurate taxi demand forecasting is essential for efficient and equitable urban transportation management. This study proposes a hybrid framework for spatio-temporal taxi demand prediction, evaluated on a large-scale dataset from Chicago. The proposed approach consists of three main components. First, missing demand data are reconstructed using a combination of Non-negative Matrix Factorization (NMF) and White Shark Optimization (WSO), enabling the recovery of latent spatial–temporal structures. Second, a Long Short-Term Memory (LSTM) network is employed to capture temporal dependencies in demand patterns. Third, a fairness-aware regularization mechanism based on Relative Hardship Proximity (RHP) is introduced to control disparities in prediction errors across demographically similar regions. This mechanism explicitly penalizes unequal error distributions, guiding the model toward more balanced performance across urban areas. Experimental results demonstrate that the proposed framework effectively balances predictive accuracy and spatial equity, achieving a Root Mean Square Error (RMSE) of 2.8 and a Mean Absolute Error (MAE) of 1.4. Crucially, the model yields up to a \(50\%\) improvement in spatial equity (Equity Gain) while incurring a minimal accuracy trade-off, reflected by a Price of Fairness (PoF) of approximately \(7\%\) . These findings demonstrate that incorporating structured fairness constraints into demand forecasting models can significantly enhance equity while maintaining high predictive performance, offering a practical and scalable solution for fair urban transportation planning.