Machine Learning - Singapore's Unable to Board (UTB) Bus Instance Prediction

Model Accuracy: 90%

Tags: #Government #Classification #Prediction #Transport #Singapore #DataVisualisation


The Land Transport Authority (LTA) spearheads land transport developments in Singapore. They plan, design, build and maintain Singapore’s land transport infrastructure and systems to strengthen Singapore’s land transport connectivity and integrate a greener and more inclusive public transport system complemented by walk and cycle options. LTA also harness technology and innovation to strengthen the rail and bus infrastructure and develop exciting options for future land transport.

The LTA continually looks at new ways to enhance the service levels and reliability of the public transport system. This includes gathering insights to understand the evolving travel needs of commuters to better plan for future of travel. As the regulator of public bus services together with the Public Transport Council, the LTA regularly monitors the quality of bus services to ensure smoother journeys and seamless connectivity for commuters.

As part of the LTA’s efforts to improve bus service performance, the LTA is looking out for a cost-effective data collection and analytics solution to accurately detect UTB instances.

Problem Statement

How can LTA monitor and detect passenger UTB instances to optimise the deployment of buses on public bus services?

Solution (Proof-of-Concept)

Analytico Asia built a Machine-Learning (ML) powered automated dashboard that shows the UTB details by day, time, bus stop, and bus-service. The solution is currently achieving a 90% accuracy in predicting UTB.

The number of UTBs for each bus instance is determined through passenger-level ticketing data. The solution models and forecasts if a passenger is UTB by predicting the time at which he/she reaches a particular bus-stop and cross-referencing against bus arrival times.

The solution’s ML models accurately predict UTB instances by using historical ticketing and surveyor data. Newly generated variables such as temperature, recurring trip, peak hour, public holiday, and school holiday among others are used to maximise the accuracy of the predictions.

With this solution, the LTA will be more equipped to improve the commuters’ experience through more effective and efficient deployment of the buses.