Skip to main content
Case Study

Passenger Numbers – Back on Track.

Jernbanedirektoratet, the Norwegian Railway Directorate, is a government agency with the strategic responsibility for the Norwegian Railway Network. It was formed on the basis of the railway reform and became operational on 1 January 2017. The directorate is responsible for developing the railway as an integrated transport system. Their mandate is to create an efficient, safe and environmentally friendly railway network benefiting passengers and operators.

Case Study Details

  • Client
    (Norwegian Railway Directorate)
  • Location
  • Year
    Since 2023

The Challenge.

Jernbanedirektoratet, as the supervisory body for the Norwegian railway system, has been working on improving its measurement accuracy of passenger numbers and the subsequent revenue distribution that follows from that. Only about 40% of all the trains within the network are equipped with passenger counters and the existing counters are not 100% reliable.

As a result, Jernbanedirektoratet has embarked on a program to create a sophisticated statistical model to accurately predict passenger numbers for all trains that addresses any false counts or random gaps. The goal was to establish a common model that can be relied upon and accessed by all operators rather than each operator developing their own disparate model/system.

The Railway Directorate set specific objectives for the statistical model to ensure fairness and credibility in the calculation of passenger numbers:

  • Comparability: The same model and method will be used for all operators, ensuring consistency and fairness in the calculation of passenger numbers.
  • Justice: The model will prevent any skewed interpretations or biases, as there will be no room, opportunity, or incentive for operators to manipulate the numbers.
  • Credibility: The statistical model will be developed and operated by an independent and competition-neutral party, ensuring the credibility and reliability of the passenger numbers compiled

Jernbanedirektoratet issued a public tender for the deployment and operation of a solution aimed at accurately predicting passenger numbers. Kogenta was selected as the supplier after an overall assessment based on price, quality, and competence.

Machine Learning at the Heart of Kogenta Predictor

The Solution.

Kogenta’s solution makes use of the Kogenta Predictor product at it’s core. A Managed Service was provided around the Kogenta Predictor product that saw Kogenta take responsibility for the ingestion of data, the creation of the statistical model and the subsequent operation of the system to deliver the necessary results.

The Kogenta Predictor product ingests a wide range of data sources that are then processed and enriched by Kogenta’s KETL engine. KETL is specifically designed to support Kogenta’s location and movement analytics solutions.

The key data sources used are:

  • APC: For each door in a train with APC-equipment installed the counter returns the number of boarding and alighting passengers on the specific departure.
  • Production data: Overview of all departures on the specific stations and departure. Includes information on trainID, planned and actual arrival and departures for all trains – both trains with APC and trains without APC equipment.
  • Weather: Using the API for to extract different weather information like temperature, precipitation, air pressure, humidity and wind.
  • Calendar information: The calendar information attributes key date information to specific dates.
  • Traffic data: This includes some variables extracted from the production data and other calculations to describe disruptions in the service.
  • Area characteristics. To describe the area the stations are situated in we extract key information from SSB.

The statistical model is built using a unified analytics platform designed to accelerate innovation by bringing data science, engineering, and business teams together. It offers collaborative workspace and tools for data engineering, data science, and machine learning workflows.

The machine learning implemented is a gradient boosting algorithm, designed for speed and performance. The algorithm used belongs to the family of gradient boosting algorithms, which sequentially adds weak learners (decision trees) to improve the model’s predictive performance. It works by minimizing a loss function, which measures the difference between actual and predicted values, through gradient descent optimization.

The model estimates passenger numbers in real time and is made available via API and a portal for the operators, Entur and the directorate to access.

“Kogenta Predictor combines the ability to collect and create the required dataset with sophisticated machine learning algorithms to model highly accurate and reliable predictions about passenger usage that is superior to traditional methods and techniques.”

Complete and Accurate Passenger Counts

The Result.

The Kogenta solution was put into live production in December 2023 and has been successfully rolled out to the largest rail operator in Norway: Vy.  Vy’s network includes local, intercity and regional trains on fifteen different train lines in eastern Norway.  Deployment to two additional operators is planned early in Q2, 2024.

The results from the first deployment have been superb with data from the model already being used by several public agencies.

Would you like to know more about this product deployment?