Thesis Work: 30 hp - Anomaly detection in credit portfolio reporting

Background:
Traton Financial Services provides financing services in over 50 countries for the Traton Group’s brands: Scania, MAN, International, and Volkswagen Truck & Bus. Each month the local entities report a set of data about the roughly 250 000 contracts that are in the portfolio at any given time. It is essential for risk management and accounting that this reporting is accurate and anomalies in this data can be leading indicators of risks arising in the portfolio.

At the moment basic sanity checks are performed on the reported data based on known scenarios, but it would be desirable to be able to detect new types of anomalies as they emerge. Unsupervised machine learning could potentially achieve this.

The main challenge is that while some combinations of values are incompatible and are easy to detect, other anomalies only become apparent when looking at a time series. For instance the value of an asset should go down over time and a customer cannot suddenly be 90 days overdue from one month to the next. Another challenge is noise in the data, for instance the value of the asset can actually go up if the local currency appreciates. There are also situations that are rare but are not an anomaly.

Objective
Create and validate a potential method for ongoing monitoring of data quality and early risk detection with unsupervised machine learning.

Job description

1. Gain an understanding of the monthly portfolio reporting data structure

2. Research suitable unsupervised machine learning algorithms to detect anomalies

3. Research ways to measure effectiveness and relative “strangeness” of the data

4. Based on the real data set create a reference test data set with anomalies inserted

5. Practically try different data transformations and machine learning algorithms and measure their ability the known types of anomalies

6. Theoretically evaluate the ability to detect unknown types of anomalies

7. Bonus: Try using a Large Language Model to explain the nature of the anomaly

The expected end result is a (python) code that performs the analysis combined with a paper describing the above mentioned aspects.

Education/program/focus

Indicate education, program or focus: Computer science

Number of students: 1

Start date for the thesis work:JANUARY 2026

Estimated time required: 20 weeks

Contact persons and supervisors:
Mikael Agerbo, Audit Data Analytics Manager, mikael.agerbo@tratonfs.com

Application:
Your application must include a CV, personal letter and transcript of grades

A background check might be conducted for this position. We are conducting interviews continuously and may close the recruitment earlier than the date specified.