Privacy-Preserving Classification of Customer Data without Loss of Accuracy

Abstract

Privacy has become an increasingly important issue in data mining. In this paper, we consider a scenario in which a data miner surveys a large number of customers to learn classification rules on their data, while the sensitive attributes of these customers need to be protected. Solutions have been proposed to address this problem using randomization techniques. Such solutions exhibit a tradeoff of accuracy and privacy: the more each customer's private information is protected, the less accurate result the miner obtains; conversely, the more accurate the result, the less privacy for the customers. In this paper, we propose a simple cryptographic approach that is efficient even in a many-customer setting, provides strong privacy for each customer, and does not lose any accuracy as the cost of privacy. Our key technical contribution is a privacy-preserving method that allows a data miner to compute frequencies of values or tuples of values in the customers' data, without revealing the privacy-sensitive part of the data. Unlike general-purpose cryptographic protocols, this method requires no interaction between customers, and each customer only needs to send a single flow of communication to the data miner. However, we are still able to ensure that nothing about the sensitive data beyond the desired frequencies is revealed to the data miner. To illustrate the power of our approach, we use our frequency mining computation to obtain a privacypreserving naive Bayes classifier learning algorithm. Initial experimental results demonstrate the practical efficiency of our solution. We also suggest some other applications of privacy-preserving frequency mining.

Privacy Preserving Frequency Mining in 2-Part Fully Distributed Setting

“…In contrast, the cryptographic approaches proposed in [6], [25] provided strong privacy without loss of accuracy. The key idea of these approaches is a private frequency computation method in the fully distributed setting that allows the miner to compute frequencies of values or tuples in the data set, while preserving privacy of each user's data.…”

“…A general definition of secure multi-party computation in the semi-honest model is stated in [8]. This definition was derived to make a simplified definition in the semi-honest model for privacy preserving data mining in the fully distributed setting scenario [6], [25]. This scenario is similar to 2PFD setting, so here we consider the possibility that some corrupted users share their data with the miner to derive the private data of the honest users, we assume that all users are semi-honest, thus any user can be corrupted.…”

“…The methods based on perturbation (e.g., [1], [3], [27]) are very efficient, but have a tradeoff between privacy and accuracy. The methods based on cryptography (e.g., [11], [23], [25]) can safely preserve privacy without loss of accuracy, but have high complexity and communication cost. These privacy preserving data mining methods have been presented for various scenarios in which the general idea is to allow mining datasets distributed across multiple parties, without disclosing each party's private data [2].…”

Accuracy in Privacy-Preserving Data Mining Using the Paradigm of Cryptographic Elections

“…In this section we briefly describe the work in [1], which is, to our best of knowledge, the first scheme that used a variant of the homomorphic election model in order to build a privacy preserving frequency mining algorithm. This algorithm is then used in [1] as a building block to design a protocol for naive Bayes learning.…”

“…To this end we argue in favor of borrowing knowledge from a broad literature dealing with cryptographic elections via the Internet. We discuss some weaknesses and describe an attack on a recent PPDM scheme of Yang, Zhong and Wright [1] which, to our best knowledge, was the first work that used a variation of the classical homomorphic model [25] for online elections. Our PPDM approach will be based on the classical homomorphic model of Cramer, Gennaro and Schoenmakers [25] for online elections, and more particularly on some recent extensions proposed in [26,27] for multi-candidate elections.…”