San Francisco, CA/USA

Increased passenger satisfaction through improved departure predictions

The Golden Gate Bridge, Highway, and Transportation District represents a unique district within the State of California. Its primary responsibility is to manage and preserve the iconic Golden Gate Bridge, along with overseeing two integrated public transportation networks: Golden Gate Transit and Golden Gate Ferry. These systems effectively link the counties of Marin, Sonoma, San Francisco, and Contra Costa, providing convenient travel options for the residents and millions of visitors to these regions.

Project at a glance

177

fixed-route vehicles

10

mil annual visitors

7

ferries

The task

Due to the challenge of a mixed urban and rural service, the District's departure prediction accuracy rates were averaging 57 percent. This meant that passengers could not fully rely on catching their bus on time, or on getting the right prediction information. The District set out to change this by improving the reliability of their bus departure predictions across all routes and subsequently increasing passenger satisfaction. 

The solution

Through a pilot project, INIT’s subsidiary, inola, implemented an artificial intelligence and machine learning (ML) software. The machine learning software provided predictions using historical data and real-time information. Independent from platforms and operating systems, it processed large amounts of data (Big Data) while MOBILEstatistics, INIT’s system for analysis and statistics, collected the operational data (e.g. GPS data) and processed historical driving times. Based on this data, various trainer systems were made available to the ML-Core. After each training session using processed historical data, the ML model was updated creating a suitable model for all bus prediction times.

ML prediction, which is different from the usual linear prediction, takes into account things like road work, accidents, or events that are planned at the last minute. Additionally, the time of day and the days of the week are also considered. The ML prediction gathered the District's newly forecasted departure times for each stop by combining individual values and sending them to different processes. The Passenger Information system displayed these live predictions, considering factors like current traffic conditions and the driving time of the preceding vehicle.

The District went from a 57 % prediction accuracy rate to an 89% prediction accuracy rate increasing reliability and improving passenger satisfaction.

The conclusion

The reliability of the prognoses has increased immensely from 57 percent prediction accuracy to 89 percent.  Passenger satisfaction has increased because they are able to plan their journeys more easily thanks to ML prediction. In San Francisco, a new era in passenger information has arrived. 

 

Privacy settings
This site uses cookies to provide you with an optimal website experience. This includes cookies that are essential to the operation of the site and cookies that you need to agree to use. You can also reach this panel later via a text link in the data protection declaration and can further adjust your cookie settings on our website to your preferences at any time. You will find additional information in the detailed view below and in our data protection declaration.
Required
These cookies are necessary to enable the basic functions of our website. We store personal page settings and your language settings.
Statistics
To further improve our website, we collect anonymized data for statistics and analysis via Matomo. With the help of these cookies we can, for example, determine the number of visitors and the effect of certain pages of our website in order to optimize our content.
Video statistics
Third-party content – e.g. Vimeo or YouTube videos – can be embedded and displayed. To further improve our communication, we collect anonymous statistical data.
Settings saved
Accept selected
Accept all