论文信息 - Research on Differentially Private Bayesian Classification Algorithm for Data Streams

Research on Differentially Private Bayesian Classification Algorithm for Data Streams

Considering the privacy leakage risk of data streams mining, a differential private Naive Bayes classification algorithm for data streams (DP-NB) is proposed in the paper. The Gaussian Bayesian model is firstly adopted and the methods of feature selection, cross-validation and weighting mechanism are used to improve the model and solve the problem of concept drift. Then noise perturbation is added to the model parameters and incremental update is presented to optimize the Gaussian classifier model. Theoretical analysis of the DP-NB algorithm proves that it achieves $\mathcal{E}-$ differential privacy. Finally, experimental results show that the DP-NB algorithm has better classification accuracy while satisfying privacy protection.

Shuyu Li | Guozheng Zhang

[1] M. C. Padma,et al. Deterministic Concept Drift Detection in Ensemble Classifier Based Data Stream Classification Process , 2019, Int. J. Grid High Perform. Comput..

[2] Cynthia Dwork,et al. Differential Privacy , 2006, ICALP.

[3] Qingfeng Wu,et al. Naive Bayes Classification Algorithm Based on Optimized Training Data , 2012 .

[4] Antti Honkela,et al. Differentially private Bayesian learning on distributed data , 2017, NIPS.

[5] Ninghui Li,et al. Differential Privacy: From Theory to Practice , 2016, Differential Privacy.

[6] David M. W. Powers,et al. Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation , 2011, ArXiv.

[7] Geoff Holmes,et al. New ensemble methods for evolving data streams , 2009, KDD.

[8] Kato Mivule,et al. Towards A Differential Privacy and Utility Preserving Machine Learning Classifier , 2012, Complex Adaptive Systems.

[9] Davide Anguita,et al. Differential privacy and generalization: Sharper bounds with applications , 2017, Pattern Recognit. Lett..

[10] J. C. Díaz,et al. An incremental gaussian mixture network for data stream classification in non-stationary environments , 2018 .