SubeYa: A System for Predicting Passenger Demand at Train Stations

Meza A.; Acuña A.; Santisteban J. · IET Intelligent Transport Systems (2026) · DOI: 10.1049/itr2.70185

Resumen

Public transportation in large cities faces significant challenges due to population growth and accelerated urbanization, leading to congestion and prolonged waiting times in systems such as line 1 of the Lima metro (EL1ML). To address this issue, the present study introduces SubeYa, a web-based platform that employs the Prophet predictive model to estimate passenger demand at specific stations based on historical data. Following a rigorous strict hold-out validation (training on 2019–2023, testing on 2024–2025) and per-station Grid Search tuning, the model achieved a global mean absolute error (MAE) of 259.10. This performance significantly outperforms robust statistical baselines, including ARIMA (MAE = 1139.19) and SARIMA (MAE = 1198.03), demonstrating the superior capability of the additive architecture to handle the structural breaks and irregular holiday patterns inherent to the network. Validation with real users revealed an average reduction of 40% in waiting times, accompanied by high satisfaction in terms of usability (4.4/5). SubeYa represents an innovative solution for travel planning, standing out for its practical and accessible implementation in the Latin American context, where similar tools are scarce.

Instituciones: Universidad Peruana de Ciencias Aplicadas