Privacy Preserving in Distributed SVM Data Mining on Vertical Partitioned Data

Abstract

Data mining algorithms tacitly quite access to the data either at centralized or distributed form. Distributed data becomes a big challenge and cannot handle by a classical analytic tool. Cloud Computing can solve the issues of processing, storing, and analyzing the data at distributing locations within the cloud. However, a significant problem that is preventing free sharing of data is privacy and security issues, therefore obstructing data mining schemes. Lately, there is increasingly hard to find a solution to these problems. Due to the existing knowledge in a more distributed data and better for data mining issues. An important task of data mining and machine learning is classification, a widely used in classification is support vector machine (SVM) algorithms applicable in many various domains. In this paper, we proposes a privacy-preserving solution for SVM classification. Our workaround constructing a global SVM classification model from vertically partitioned distributed data at multi-parties based on Gram matrix, without revealing a party's data. We proposed an efficient and preserve privacy protocol for SVM classification on vertical partitioned data. Our experimental results, the accuracy of distributed SVM using Gram matrix up to 90% and the privacy not compromised.