FORECASTING BOARDING PASSENGER COUNT USING LONG SHORT-TERM MEMORY (LSTM)NETWORKS

To run smoothly, city transport systems must forecast boarding passenger flow from a station. This aids resource management, scheduling, and passenger comfort. However, forecasting abnormal passenger flow—such as sudden passenger increases or decreases—is difficult. These abnormalities are rare, thus there is little historical data to learn from. Deep learning models must handle weekdays, weekends, holidays, and timeslots. LSTM neural network model is used to solve this problem. LSTM neural networks retain long-term dependencies in sequential data, making them ideal for time series prediction. They need good training to recognize anomalous passenger flow for boarding. This uses data patterns including daily, weekly, and monthly cycles to predict. Periodic features help the model predict passenger flow. For the same, we have gathered data received from e-ticket machines for the month of January 2022 to November 2023 pertaining to Thane Municipal Transport system. In this study we have documented four various types (i.e. huge demand to minimal demand) of boarding stations with detail analysis. The LSTM model is initially trained with hyper-parameter tuning on the entire passenger flow data, including normal and anomalous patterns related to time-slots, holidays and day of week. The model learns general trends and periodic fluctuations. December 2023 passenger hourly boarding counts have been forecasted. Such models have been evaluated using RMSE, MAE, R-Square, and necessary graphical representations. The outcome of evaluation parameters suggests that models are accurate and compatible for deployment and results of testing data is in-line with real count. In summary, the time-series models efficiently handles the difficulty of anticipating irregular passenger flow by using the capabilities of LSTM approaches, together with a clever combination strategy based on real-time fluctuations. This improves passenger boarding forecasting, which are essential for intelligent city transport systems.