Preserving Private Knowledge In Decision Tree Learning

Abstract

Data mining over multiple data sources has become an important practical problem with applications in different areas. Although the data sources are willing to mine the union of their data, they don’t want to reveal any sensitive and private information to other sources due to competition or legal concerns. In this paper, we consider two scenarios where data are vertically or horizontally partitioned over more than two parties. We focus on the classification problem, and present novel privacy preserving decision tree learning methods. Theoretical analysis and experiment results show that these methods can provide good capability of privacy preserving, accuracy and efficiency.