Bus travel time prediction pilot outperforms existing service by 74% with Azure Machine Learning.

TransLink is Metro Vancouver’s transportation network. Their footprint spans over 1,800 square kilometers and serves residents and visitors with trip planning and public transit through their bus, train, and ferry services.

With the advent of affordable transportation disruptors, they had to innovate and adapt quickly to keep up with customer expectations. Traditionally, the number of riders, and revenue produced from those riders, had been paramount to TransLink’s bottom line. They realized they needed to change their way of thinking by taking a more human-centered approach to measuring success. That meant recognizing that the ridership experience was key to retaining their rider base and attracting new customers.

Through this realization, they were able to identify a key area of improvement: reliable and accurate predictions of bus departure times at bus stops along its routes.

The Challenge

The customer had a long-term solution in place but wanted to measure and identify areas of improvement. A modelling proof-of-concept delivered by Microsoft’s Data Science team led to a referral for TransLink to engage MNP as a preferred implementation partner for the delivery of a replacement prediction service.

The Approach

The client selected 13 routes to audit across their transit networks with a goal of delivering a solution that would out-perform the information produced from their existing tool. The success measures were stated and modest, with 70% improvement as the ultimate goal.

Through a machine-learning model, MNP assisted in delivering:

  • Integration services to provide access and quality controls for weather
  • Built and trained a machine learning model that revised algorithms and modelling approach to reduce, on average, 70% of existing service predictions that had greater than (5) minutes error, to less than (5) minutes error
  • Training and knowledge transfer to TransLink

The Result

The results were beyond everyone’s expectations, the pilot outperformed the existing solution by reducing errors by 74%.

The team proved that existing data assets can be leveraged to add tremendous value while also revealing new opportunities for increased optimization. This led to an improved rider experience, with increased reliability and consistency in the TransLink bus services.