An efficient and practical approach for privacy-preserving Naive Bayes classification

Nowadays, the development of machine learning has brought about tremendous benefits. Nevertheless, the process of building machine learning models can violate sensitive and private information in data, especially in some specific domains such as banking, medical, e-commerce. In this paper, we propose a novel privacy preservation solution for the Naive Bayes classification technique, a powerful machine learning algorithm. Because of being based on a new cryptographic protocol called secure multi-sum computation, the proposed solution not only protects data holders’ privacy but also guarantees the classification model’s accuracy. Moreover, the experimental results demonstrate that our solution outperforms the typical privacy-preserving Naive Bayes classifiers, and it is efficient in real-life applications. • A new secure multi-sum computation protocol is proposed for privately computing many sum values in one round of computation. • A novel and efficient privacy-preserving Naive Bayes classifier based on the secure multi-sum computation protocol is propounded. • The new privacy-preserving Naive Bayes classification solution has high performance when tested on the SMS spam dataset with privacy constraints.