Outsourcing Data Mining Tasks to Cloud while Preserving Customer privacy

DOI: 10.5176/978-981-08-6308-1_59

Authors: Yi-Ming Chen, Yan-Hao Chu

Abstract:

Recently, as more and more enterprises are likely to outsourcing time consuming data mining tasks to the cloud for saving both time and money, it becomes important to ensure the customer’s privacy because the decentralized nature of cloud services increases the risk of data leakage. Traditional approaches either prevent the edge mining server from contacting the original customer data or encrypting the queries to protect the customer’s knowledge behind the query logs. Most of the approaches, however, still need to put the huge volume of original customer data in a central data server which may be attacked by malicious users. In this paper, we propose a BKM approach to solve this problem. By the BKM approach, before put to the cloud, all of the customer’s data are transferred to a string of is and 0s by bloom filters. Then these seems non-meaningful strings are grouped by K-means method. Finally we use the MFTS (Maximal Frequent Term Set) algorithm to find the most representative pattern in each group and feedback these patterns to customers. To illustrate the applications of BKM approach, we use a set of security alerts adopted from the DARPA dataset to show how the BKM mining can help customers to find out the statistics of the attack patterns while revealing none of customer’s privacy in cloud.