Publication Date
12-2019
Date of Final Oral Examination (Defense)
8-8-2019
Type of Culminating Activity
Thesis
Degree Title
Master of Science in Computer Science
Department
Computer Science
Supervisory Committee Chair
Gaby Dagher, Ph.D.
Supervisory Committee Member
Bogdan Dit, Ph.D.
Supervisory Committee Member
Jyh-Haw Yeh, Ph.D.
Abstract
Privacy-preserving distributed data mining is the study of mining on distributed data—owned by multiple data owners—in a non-secure environment, where the mining protocol does not reveal any sensitive information to the data owners, the individual privacy is preserved, and the output mining model is practically useful. In this thesis, we propose a secure two-party protocol for building a privacy-preserving decision tree classifier over distributed data using differential privacy. We utilize secure multiparty computation to ensure that the protocol is privacy-preserving. Our algorithm also utilizes parallel and sequential compositions, and applies distributed exponential mechanism to ensure that the output is differentially-private. We implemented our protocol in a distributed environment on real-life data, and the experimental results show that the protocol produces decision tree classifiers with high utility while being reasonably efficient and scalable.
DOI
10.18122/td/1604/boisestate
Recommended Citation
Kumar, Manish, "Secure Two-Party Protocol for Privacy-Preserving Classification via Differential Privacy" (2019). Boise State University Theses and Dissertations. 1604.
10.18122/td/1604/boisestate