LSTM Network Integrated with Particle Filter for Predicting the Bus Passenger Traffic

Area of Science:

• Artificial Intelligence
• Data Science
• Transportation Engineering

Background:

• Accurate passenger traffic prediction is crucial for efficient public transportation scheduling.
• Traditional methods often struggle with the complex temporal and spatial dynamics of traffic data.
• Integrating advanced machine learning with statistical filtering offers a promising approach.

Purpose of the Study:

• To develop and validate an integrated model combining deep learning and Bayesian filtering for passenger traffic prediction.
• To analyze temporal patterns in passenger traffic data.
• To enhance bus scheduling efficiency through accurate short-term forecasts.

Main Methods:

• Utilized a Long Short-Term Memory (LSTM) network for time series sequential prediction.
• Integrated a Particle Filter (Bayesian filtering) to capture Markovian behavior.
• Analyzed temporal (morning, noon, post-noon) and spatial features of traffic data.
• Statistically modeled identified temporal patterns.

Main Results:

• The integrated model accurately predicted passenger flow for the next thirty days.
• Identified distinct morning, noon, and post-noon traffic patterns.
• Achieved a high coefficient of determination (R²) of 0.88 in predictions.
• Demonstrated effectiveness even with small training datasets.

Conclusions:

• The proposed deep learning and Bayesian filtering model is effective for passenger traffic prediction.
• The model's ability to capture temporal dynamics enhances its predictive power.
• Accurate short-term forecasts facilitate optimized bus scheduling and resource allocation.