Security & Fraud

The application and development of machine learning methods in the field of (network) security and fraud is an active field of research in the Data Science Chair. In the DeepScan project, we are developing methods to detect anomalies, ICT security incidents and fraudulent behaviour in business software. Other research projects are currently working on the detection of security incidents in corporate networks or on application layer.

The following staff member have open topics for practica, bachelor and master theses:

Fraud-Detection, Machine Learning for Computer-Security	Daniel Schlör
Explainable AI, Fraud-Detection	Julian Tritscher

In the case of excellent performance there is also the chance to submit the thesis as an article to a computer science conference and to be co-author on a scientific publication early in your studies!

Open Topics:

Detecting Fraud in Company Data

Modern companies record large amounts of data regarding their daily workflow. While professional auditors regularly screen this data to find fraudulent abuse of company assets, the size of tracked data in companies is continously increasing. This makes automated fraud detection through artificial intelligence a promising research topic.
This work focuses on how well machine learning approaches can detect fraud in company data. Tasks can range from applying machine learning to different views on the company workflow, to evaluating the use of different types of machine learning approaches.

Supervisor: Julian Tritscher

Understanding Neural Network Reasoning When Detecting Anomalies

Neural networks offer strong performance in many application areas, but their reasoning is difficult to follow due to their complexity. Through the desire to understand the decision process of complex neural networks, explainable artificial intelligence (XAI) has recently become a popular research topic. While many approaches exist for explaining neural network decisions, they are usually designed for popular tasks such as image classification.
This work focusses on investigating the ability of existing XAI approaches to explain neural networks in anomaly detection, a specialized domain where small numbers of anomalies need to be separated from large amounts of normal data. The topic can be taken in different directions, such as evaluating different XAI approaches or improving explanations of specific XAIs.

Supervisor: Julian Tritscher

Agent-based Simulation of Business Processes

Für viele Anwendungen, wie Fraud Detection oder Process Monitoring werden Log-Daten von Geschäftsprozessen benötigt, um Modelle zu trainieren. Obwohl diese in modernen ERP Systemen automatisch aufgezeichnet und ausgewertet werden, und damit leicht zugänglich wären, gibt es kaum frei verfügbare Datensätze. Ausgehend von stark aggregierten Daten soll im Rahmen dieser Arbeit ein Simulationssystem entwickelt werden, das Geschäftsprozesse und deren Dokumentation simuliert und dadurch synthetische Daten generiert, die einer Abbildung realer Prozesse möglichst nahe kommen.

Betreuer: Daniel Schlör

Anomaly Detection and the modelling of normality

Unter Anomaly Detection verstehen sich Data-Mining Methoden um seltene Ereignisse (Anomalien) zu finden, die sich vom Großteil der Daten unterscheiden.
Mögliche Anwendungsgebiete sind:

(Network) Security
Fraud Detection
Fault Diagnosis
Novelty Detecting im Bereich Text-Mining

Neben dem Finden von Anomalien ist die Modellierung des Normalzustands eine wichtige Teilaufgabe. Im Rahmen dieser Arbeit sollen verschiedene Verfahren zur Anomaly Detection und der Modellierung des Normalzustands in Bezug auf ein oder mehrere Anwendungsgebiete verglichen werden.