Differentially Private Naïve Bayes Classifier Using Smooth Sensitivity



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 talk (1) I will give background on Smooth Sensitivity defined by Nissim, Raskhodnikova, and Smith. (2) I will present our recent work in which 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, 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 Differential Privacy.



Farzad Zafarani, PhD student

Purdue University,
West Lafayette,



Farzad Zafarani is a PhD student at Purdue University, under the supervision of Chris Clifton. Previously, he did his bachelor's and master's studies at Isfahan University of Technology and Oregon State University, respectively. His research encompasses differentially private machine learning and lower bounds in differential privacy.