A Bayesian Additive Model for Understanding Public Transport Usage in Special Events

Area of Science:

• Computational Social Science
• Transportation Engineering
• Data Science

Background:

• Public special events (sports, concerts, festivals) cause unpredictable transportation disruptions.
• Current methods for predicting event-related travel rely on manual processes and personal experience, proving inefficient and costly.
• Concurrent events exacerbate prediction difficulties for transportation operators.

Purpose of the Study:

• To develop a data-driven model for predicting public transportation demand in special event areas.
• To enable a more adaptive transportation system by accurately forecasting travel patterns.
• To disaggregate trip counts in scenarios with multiple concurrent events.

Main Methods:

• A Bayesian additive model incorporating Gaussian process components was developed.
• The model integrates smart card records from public transport with event information extracted from the web.
• An efficient approximate inference algorithm using expectation propagation was employed for prediction and disaggregation.

Main Results:

• The proposed model accurately predicts total public transportation trips to special event areas.
• The algorithm successfully disaggregates gross trip counts into event-specific and routine travel components for concurrent events.
• The model demonstrated superior performance, outperforming the best baseline model by up to 26% in R2 using real-world data from Singapore.

Conclusions:

• The developed Bayesian model offers a robust and efficient solution for predicting transportation demand influenced by public special events.
• The model's ability to handle concurrent events and provide component-wise insights enhances transportation system adaptability.
• This approach provides valuable explanatory power for individual model components, aiding in understanding travel behavior drivers.