Modeling Short-Term Passenger Flows in Metro and Bus Systems Using Meteorological Data: Deep Learning Model Comparisons

In this study, a Long Short-Term Memory (LSTM) model with extra variables such as weather conditions and school days was developed within a multi-scale framework in order to forecast passenger flow in both bus and rail systems, covering both regional and route-level analyses. In addition, the performance of the LSTM model was compared against three separate deep learning models. Among these, the Nonlinear Autoregressive Network with Exogenous Inputs (NARX) time series model produced the lowest error values, achieving a high level of accuracy. While no considerable changes were observed in regional rail passenger flow as a result of the inclusion of weather-related variables, a 2.2% drop in the RMSE value was achieved in bus passenger flow at the regional level; however, this improvement remains relatively modest. In contrast, at the route level, RMSE values declined by 2.4% for rail and 3.69% for bus routes. These findings reveal that the inclusion of weather-related variables significantly improves the prediction of bus passenger flow, underlining the benefits of integrating such data into forecasting models. Furthermore, the findings of this study analytically support transportation planners in making more informed, data-driven decisions regarding scheduling and capacity management.