Urban demand data exhibit both spatial proximity and temporal variability, requiring an analytical framework that jointly captures spatial correlation and temporal continuity. This paper proposes a Spatiotemporal Quantile Regression model with K-Neare...
Urban demand data exhibit both spatial proximity and temporal variability, requiring an analytical framework that jointly captures spatial correlation and temporal continuity. This paper proposes a Spatiotemporal Quantile Regression model with K-Nearest Neighbor (KNN) Fused LASSO, combining the robustness of quantile regression with the structural regularization of fused LASSO. The model incorporates fused penalties to estimate spatial and temporal smoothness simultaneously and employs the pinball loss for robust estimation under outliers and asymmetric noise. Parameter estimation is efficiently performed via the Alternating Direction Method of Multipliers (ADMM), enabling scalability to large datasets. The proposed approach achieves superior robustness and boundary recovery compared to mean regression and spatial-only models, particularly at extreme quantile levels, and automatically identifies interpretable spatiotemporal clusters in heterogeneous regions. These results show that the proposed model provides enhanced stability and structural recovery over existing methods.