Improving tram operations through data fusion and machine learning


5 Minute read

A project to better estimate passenger loads and predict real-time crowding for Melbourne trams is concluding, with the outcomes to improve the overall efficiency and operation of tram services.

Detailed knowledge of how public transport services are used and by how many people is the basis for design, operation and adjustment of service. While advancement in sensing technologies enable transit operators to monitor the variabilities in passenger flows continuously and consistently, the data
from automatic sensors is also incomplete, inaccurate and biased, posing a challenge to this process.

For example, they don’t account for fare evasion or, specific to Melbourne, the free tram zone in the Central Business District (CBD), where passengers are actively discouraged to tap on or off tram services.


In the context of Melbourne’s rapidly evolving demand for tram services – due to the pandemic altering the day-to-day lives of Australians and their travel patterns, as well as an increase in modes of transport (including bikes, scooters, along with buses and trains) – having accurate data to inform
efficient services is particularly relevant.

This is what the University of Melbourne, in collaboration with the Victorian Department of Transport, Yarra Trams and Cubic Transportation Systems is working on through a data fusion and machine learning project, taking place in Melbourne.

The project is part of the Australian Integrated Multimodal Ecosystem (AIMES), a world-first living laboratory testing highly integrated transport technology with a goal to deliver safer, cleaner and more sustainable urban transport outcomes.

Led by Dr Neema Nassir and Professor Majid Sarvi, along with their team of researchers and students, the project brings together automatically and passively collected data (such as passenger counts, pedestrian sensors, and fare card transactions) to enable more accurate estimates of service load and
predict crowding levels in real-time.


By using data fusion techniques over conventional fare and data collection, space, time and population are all captured; alternative and independent sources of data (from automatic passenger counts to on-board footage) complement fare card transactions. The approach is also supported by a growing body
of literature on using supervised learning models with direct passenger counts from historical observations.

Trip chaining (combining commuters’ sequential stops into a tour trajectory), network modelling and Machine Learning techniques (including deep learning and reinforcement learning) are also key aspects of the project.

Ultimately, a dynamic and comprehensive method to capture data would improve everything from user planning to operations, helping passengers choose better paths and make COVID-safe decisions. The data could also reduce crowding by informing operational interventions such as procurement of supplementary
and replacement services, to operation of tram routes with other modes of transport.


Dr Nassir, Senior Lecturer in Transport Engineering at the Department of Infrastructure Engineering and member of the AIMES team, says, the potential to improve tram operations and service efficiency is huge.

“Imagine if the traffic signal control systems at intersections could tell how many people were on board every arriving vehicle, especially trams and buses. Imagine if the traffic signal could take all that information into account when allocating green times to trams, buses, cars, bicycles, and pedestrians. This would mean better allocation of right-of-way to high-occupancy and sustainable modes of transport and more equitable traffic signal operation approach.”

With the project close to concluding, a way to reliably monitor – and respond to – day-to-day variability of travel demand and passenger responses to service disruptions, special events, restrictions (such as COVID-19) and operational interventions is that much closer.

Get in touch with the AIMES team to see how a partnership can help your business prepare for the future of traffic today.

Learn more about working with the University

This article was first published on 27 May 2022 by the Faculty of Engineering and IT.

First published on 2 June 2022.

Share this article