Assessing the Impact of Privacy-Preserving Machine Learning and Bias Introduction on Data Anonymisation

This study investigates the impact of Privacy-Preserving Machine Learning (PPML) techniques on data anonymisation, with a focus on the risk of re-identification. The key PPML method used is differential privacy, while bias is introduced through Laplace Noise. Using the PreSS Dataset, we apply a decision tree algorithm to predict the success rates of first-year computer science students. The analysis is conducted across multiple scenarios, including the original dataset, differential privacy, and Laplace Noise, with an emphasis on how these techniques influence both anonymisation and model performance. Our findings reveal the trade-offs between preserving privacy and maintaining model accuracy, offering valuable insights into the development of ethical and privacy-aware AI systems.