Ridership forecasting

This model predicts the probability of number of passenger onboarding and offboarding at each stop. This is computed as a risk assessment - what is the probability of 1 person onboarding at stop A, the probability of 2 people onboarding at A etc.

There are many examples of classification models that have been built for communicating onboard crowding in trip planners, this is a different approach where focus is on providing actionable information to traffic production. The idea is that if we can predict demand then we can make more effective use of the resources at hand - vehicles and staff. i

This enables new capabilities in public transport

• Proactive management of crowding onboard - automate action when acceptable threshold is exceeded

• Proactive trip planning - suggest alternative route when risk is high for no available seat or standing

• More effective resource allocation - balance supply (onboard capacity) and demand (net ridership)

How supply and demand can be computed

Supply

• Driving times on adjacent buses explain bus movements and enable a forecast of next week's driving times including delays

• This is then matched to the vehicle type's total on-board capacity or the desired capcity level

Demand

• Historical data (18 months) of passenger counting data

• Adjust using the most recent 4 weeks of passenger counting data

• Previous weekday is weighted up

• Adjustments using hourly estimates from passenger counting earlier the same day

• Adjustments using the delay forecasting models