NS and CQM make passenger forecasts consistent and transparent

Predicting passenger numbers is crucial

For NS, predicting the number of train passengers is vital for ensuring high customer satisfaction. These predictions determine rolling stock allocation, which in turn affects transport costs and passenger comfort specifically, ensuring enough passengers have a seat. Detailed passenger forecasts enable an optimal balance between rolling stock deployment and passenger comfort, and consequently, between affordability and customer satisfaction. But what happens when a prediction for an individual train suddenly deviates from expectations? For instance, if the model predicts higher passenger numbers for a specific train during the typically quiet month of June than during the busy month of September? NS needs to be able to explain such anomalies something the previous system could not do. NS therefore decided to simplify the system. The project was named Borealis. The exciting challenge: delivering the first forecasts based on a new, transparent methodology within just a few weeks.

Data science with big data elements

This is a classic data science project involving big data elements. Passenger number forecasts must be generated for every individual train on every specific route in the Netherlands, at any time of the year. A single ‘delivery’ comprises no fewer than 125,000 predicted combinations of train, route, and day. The input data is even more extensive: nearly 200 million data points from a wide variety of sources. An increasingly important source is the check-in and check-out data from public transport smart cards (OV-chipkaart). Another source consists of actual passenger counts conducted annually on all trains. Timetable data is also utilized. The initial delivery wasn't perfect, but it was adequate. The most important thing was that everyone was confident Borealis was the right approach precisely because we now know exactly which data and models are being used, allowing us to understand the causes of any discrepancies.

New forecasting method represents a major leap forward

NS project manager Raoul Klein Kranenbarg considers the new forecasting method a significant improvement for both passengers and NS: “First of all, because the system now makes optimal use of data from the public transport smart card (OV-chipkaart) anonymously, of course. Since checking in and out with the smart card became the standard, it has been our most important data source. It is therefore great that we have spent recent years gearing the Borealis model toward this data to continuously improve our forecasts. The model now also incorporates passenger reports submitted via the NS app and to Customer Service, which helps us identify overcrowded trains. Borealis also allows us to track empty trains. But above all, the entire system now offers much greater insight and provides consistency over time.” The plan is for NS to bring the operation of Borealis fully in-house later this year, with guidance from CQM.

_{Making texts/images public and/or reproducing them is permitted only with the express consent of CQM.
Photo credits: CQM and NS.}