| description abstract | Prediction of the extra-peak hourly ridership (EPHR) is directly related to the capacity design of subway station service facilities. In the traditional station-level EPHR prediction process, the predicted value is simply the result of the multiplication of the predicted peak hourly ridership (PHR) value by a unified extra-peak hour factor (EPHF). However, the station-level EPHR predicted by this method may be underestimated because the PHR prediction results are extracted from a line-level prediction value, rather than the station-level value. Moreover, while the existing EPHF is always determined by China’s Code for Design of Metro, it is too simple and unrefined to be applicable. The proposed station-level EPHR prediction approach exhibits significantly improved accuracy and applicability via the introduction of a least absolute shrinkage and selection operator (LASSO)-based feature selection method. The historical ridership and related attribute data of the stations are used to construct relationship models for the peak deviation coefficient (PDC) and the EPHF to make the model more explanatory. As a case study, this approach was evaluated on a real-world, large-scale passenger flow dataset from Xi’an, China, and compared with the results of the traditional method. The results indicate that the EPHR prediction accuracies of 10% to 51% of the stations are improved and the corresponding mean absolute percentage error (MAPE) is reduced by 6%–30%, as compared with the traditional method, suggesting wider applicability and higher precision for station-level prediction. A supplementary comparison with two other feature selection methods further verifies that the LASSO-based approach exhibits higher accuracy and applicability. | |
