Short-Term Passenger Flow Prediction for Subway Based on Bi-LSTM and Random Forest

Accurate passenger flow prediction is crucial for optimizing resources and enhancing customer experience in the public service industry. This paper uses cleaned and engineered historical passenger flow data for model training. The primary method is the Bidirectional Long Short-Term Memory (Bi-LSTM) model, which captures long-term dependencies in time series data. Random Forest is also used for feature regression to improve performance. The experiment evaluates four models: Long Short-Term Memory (LSTM), Fully Connected Long Short-Term Memory (FC-LSTM), Bidirectional Long Short-Term Memory (Bi-LSTM), and Bidirectional Long Short-Term Memory-Random Forest (Bi-LSTM-Random-Forest). The Bi-LSTM-Random-Forest model achieved a Root Mean Square Error of 36.24 and a Mean Absolute Percentage Error of 25.62%, outperforming the others. The results show that combining deep learning and traditional machine learning methods enhances passenger flow prediction, which is vital for business decisions and resource optimization.