Performance Analysis of Privacy Preserving Naïve Bayes Classifiers for Distributed Databases

The problem of secure and fast distributed classification is an important one. The main focus of the paper is on privacy preserving distributed classification rule mining. This research paper addresses the performance analysis of privacy preserving Naïve Bayes classifiers for horizontal and vertical partitioned databases. The Naïve Bayes classifier is a simple but efficient baseline classifier. We compare the performance of our two proposed privacy preserving Naïve Bayes protocols with basic Naïve Bayes classifier (NBC). First protocol used Un-trusted Third Party (UTP) for privacy preserving Naïve Bayes classifier for horizontally partitioned data and second protocol used secure multiplication protocol for privacy preserving Naïve Bayes classifier for vertically partitioned data. The results analysis shows that our protocols execution time is less than the existing NBC execution time since in our protocol, all parties individually calculate their probability or model parameters as an intermediate result and transfer only these intermediate results for further calculations. Accuracy of test data is same because calculated model parameters of training data are same. Our protocols are very easy to follow, understand with minimum efforts, secure and fast.