Exjobb: Fraud Detection using Graph Structures with Data Programming Methods

We believe in a society shaped by successful companies, where entrepreneurship and public benefit go hand in hand. Being part of Fortnox means being part of constant change – change where you decide how far you want to go. Our main focus is to give you the right conditions so that you can live your passion with us. Where you come from doesn't matter – we're interested in where you're going. Just as we couldn't have dreamed of the Fortnox that exists today yesterday, you can help us shape who we will be tomorrow.

Background As one of Sweden's leading players in business systems, Fortnox plays an important role in Sweden's financial and payment sector. This central role comes with the risk of the platform being misused for fraud. Traditional detection methods, such as rule-based systems and analysis of individual companies, are resource-intensive and often fail to identify complex, interconnected networks of illicit activity. These methods struggle to grasp the big picture and miss the subtle connections between seemingly unrelated transactions and entities.

The data available through a business system can be used to model a scenario as a graph: companies and people become nodes, connected by edges representing transactions, relationships, ownership, etc. This data modeling strategy captures the complex interdependencies of modern businesses, allowing for visualization and further analysis. By leveraging a graph structure, we enable the application of graph algorithms, both classic and machine learning-based, to detect patterns and connections indicating a high risk of fraud.

Therefore, we offer a thesis project focused on exploring methods for fraud detection in a graph structure using a data programming approach, including the method of labeling.

Research Goals The main goal of this thesis project is to demonstrate how graph algorithms (e.g., GNN, PageRank) can be used in conjunction with heuristic signals to detect complex fraud networks. Specific research goals include:

• Utilize the data programming method (weak supervision) to translate business logic and domain knowledge into risk signals within the graph structure.
• Evaluate and implement graph algorithms, both classic and machine learning-based, to propagate these weak signals across the network, ultimately producing a ranked "risk score" for entities (similar to a recommendation system).
• Develop metrics and an evaluation framework to compare the performance of different graph-based anomaly detection methods against traditional rule-based baselines.

Qualifications We are looking for students who:

• Are pursuing a Master's degree in Computer Science or a related field.
• Have a deep interest in Machine Learning, AI, and Finance.
• Possess good programming skills.
• Can work both independently and as part of a development team.
• Are fluent in Swedish.

Location At Fortnox's office in the heart of Linköping.

Scope 1-2 students, 30 credits each.

Fortnox – At the heart of business life Fortnox helps companies start, grow, and develop. With smart technology solutions for companies and accounting firms, accessible financing, and an ecosystem of strong partnerships, Fortnox is the center of entrepreneurship in Sweden. We are driven by a vision of a thriving society shaped by successful companies.

Fortnox was founded in 2001, with its headquarters in Växjö. Learn more at fortnoxgroup.com.

Welcome with your application!

Personal Data Processing We at Fortnox care about your personal privacy. When you submit your application to us, we will process your personal data to manage the recruitment process. The data controller for the processing of personal data is the company you apply to, and this can be any company within the Fortnox group. You can read the full privacy notice at the following link: