Differentially Private Naïve Bayes Classifier Using Smooth Sensitivity
Authors: Farzad Zafarani (Department of Computer Science, Purdue University, USA.), Chris Clifton (Department of Computer Science, Purdue University, USA.)
Volume: 2021
Issue: 4
Pages: 406–419
DOI: https://doi.org/10.2478/popets-2021-0077
Abstract: There is increasing awareness of the need to protect individual privacy in the training data used to develop machine learning models. Differential Privacy is a strong concept of protecting individuals. Naïve Bayes is a popular machine learning algorithm, used as a baseline for many tasks. In this work, we have provided a differentially private Naïve Bayes classifier that adds noise proportional to the smooth sensitivity of its parameters. We compare our results to Vaidya, Shafiq, Basu, and Hong [1] which scales noise to the global sensitivity of the parameters. Our experimental results on real-world datasets show that smooth sensitivity significantly improves accuracy while still guaranteeing ε-differential privacy.
Keywords: differential privacy, Naïve Bayes classifier, privacy, data mining, smooth sensitivity
Copyright in PoPETs articles are held by their authors. This article is published under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 license.
