Balancing Privacy and Accuracy in Machine Learning Models with Differential Privacy
Authors
Term
4. semester
Education
Publication year
2025
Submitted on
2025-06-04
Pages
89
Abstract
This thesis studies how to keep personal data safe in machine learning by using Differential Privacy (DP). It tests several models Logistic Regression, Decision Trees, Naive Bayes, and Neural Networks on the Adult Income dataset. The models are trained on both original and DP protected data with different privacy budgets. Naive Bayes works well with DP because it is sim- ple and uses probability. Ensemble mod- els also keep good accuracy across privacy levels. Neural Networks with DP-SGD bal- ance accuracy and privacy, helping reduce privacy attacks. The study suggests using Naive Bayes, ensemble models, and DP- SGD for real world use where both privacy and accuracy matter.
Documents