Differential Privacy Preserving Naive Bayes Classification via Wavelet Transform
Wenru Tang, Yihui Zhou, Mingshuang Li, Laifeng Lu · 2020
While data mining brings convenience to people, it also leads to the leakage of users' privacy. It is very meaningful to mine data while protecting user data privacy. The existing Naive Bayes classification algorithms based on differential privacy have low utility in classifying high-dimensional datasets. To solve this problem, we propose a differential privacy preserving Naive Bayes classification algorithm via wavelet transform. We perform wavelet transform on the original dataset. By retaining the approximate coefficients after wavelet transform, the purpose of data dimensionality reduction is achieved. Laplace noise is added to the dimensionality-reduced dataset, and then perform Naive Bayes classification on the noisy dataset. Experimental results show that the classification accuracy of our proposed algorithm for high-dimensional datasets is significantly better than the existing differential private Naive Bayes classification algorithm.