Modeling Rail Ridership using Long Short-Term Memory and Gated Recurrent Unit: A Case Study in Malaysia

Tok Chia Wen, Wong Zi Ming, Nor Azuana Ramli, Nur Haizum Abd Rahman, Wan Nur Syahidah Wan Yusoff, Mohd Azri Rosli, Nadzri Hj. Yusof
dx.doi.org/10.17576/apjitm-2026-1501-11

Abstract

Efficient management of rail transportation systems is important for achieving sustainable urban mobility. Identifying the factors influencing rail ridership and accurately forecasting future ridership patterns are critical for optimizing operational strategies, infrastructure planning, and resource allocation. This study investigates the primary determinants of rail ridership and develops predictive methods to forecast ridership trends. The methodology uses multivariate linear regression to identify significant predictors, along with deep learning approaches such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models for multivariate time series forecasting. The LSTM model obtained a mean absolute percentage error (MAPE) of 21.32%, an R-squared (R?) of 0.74, and a root mean squared error (RMSE) of 1,009.03; whereas the GRU model obtained a MAPE of 17.36%, an R? of 0.83, and an RMSE of 803.09. Both models effectively captured the complex patterns in the ridership data. However, the GRU model achieved a slightly more accurate result than the LSTM model. The successful use of deep learning models in this research study indicates that they may provide a strong method for representing the complexities of ridership data and identifying valuable areas for public transportation systems' research and future operational improvements.

keyword

deep learning; multivariate linear regression; rail ridership; predictive modeling.

Area

Data Mining and Optimization