Supervisor

• Malte Josten, M.Sc.

Thesis: Master’s Thesis

Detecting spam and phishing requires large amounts of training data, but real emails often contain sensitive information, making their use legally and ethically problematic from a privacy perspective. This thesis explores how synthetic data generation (e.g., using Generative AI or classical techniques) can be used to create realistic yet anonymized “ham” emails (legitimate emails). The goal: Improve spam detection models without compromising real user data.

What Will You Do?

• Generate synthetic data: Experiment with techniques like GANs, LLMs, or rule-based approaches to create realistic emails.
• Ensure privacy compliance: Develop or apply methods to guarantee that generated data cannot be traced back to real users (e.g., through differential privacy or anonymization).
• Train and evaluate models: Compare the performance of spam classifiers trained on synthetic vs. real data.
• Assess practical applicability: Analyze whether synthetic data can serve as a valid alternative to sensitive training datasets.

Prerequisities

Required

• Basic understanding of machine learning and articifial intelligence, e.g., Autoencoders or GANs (finished the course Foundations of Artificial Intelligence)
• Familiarity with Natural Language Processing techniques
• Proficiency in at least one programming language (preferably Python)
• Proficiency in using LaTeX

Optional

• You took the following courses:
  • Internettechnologies & Web Engineering
  • Advanced Methods of Machine Learning
  • Security in Communication Networks
• Familiarity with evaluation metrics for AI models
• Basic knowledge of privacy-preserving techniques
• Basic knowledge of principles related to spam and phishing