Modelado comparativo de inferencia difusa para la estimación del dwell time en el transporte para personas con movilidad limitada

Resumen

The estimation of dwell time associated with boarding and alighting events is a relevant factor in the planning and scheduling of transport services for passengers with limited mobility, as it directly affects time feasibility and route performance. This study presents a comparative analysis of two fuzzy inference approaches—Mamdani and Sugeno Type 1—for modeling dwell time under uncertain operational conditions. A dataset of 250 observations was collected through direct field measurements at healthcare facilities in Apizaco, Tlaxcala, Mexico, considering four categories of users according to the mobility aid employed: no aid, cane, walker, and wheelchair. A hierarchical modeling structure was adopted to segment the estimation process by category, while a common input variable, defined as a mobility index in the continuous range [1,10], was used for both inference systems. The Mamdani model employs linguistic output sets with centroid defuzzification, whereas the Sugeno model uses first-order functional consequents calibrated through weighted least squares. Model performance was evaluated using RMSE, MAE, MAPE, and the coefficient of determination. Results show that the Mamdani model achieved coefficients of determination in the range of approximately 0.85–0.93, while the Sugeno Type 1 model consistently obtained higher values, above 0.98 across all categories, along with substantially lower estimation errors. These findings indicate that fuzzy inference systems are suitable for dwell time estimation and that the Sugeno formulation provides advantages when accurate numerical predictions are required for integration into route planning frameworks.