Today, the volume of data collected everyday has tremendously increased. Data mining, the technology that extracts useful information and insights from the data becomes intriguing for the end users. Due to the fact that the data owner may not possess sufficient resources to perform data mining computations by his/her own, there is a need for outsourcing data mining computations to a third-party computationally powerful data mining service provider. With the advance of the Cloud computing technology, a paradigm called the Data-Mining-as-a-Service (DMaS) has emerged. Although the DMaS paradigm provides an affordable data mining solution for the data owner, there are several security concerns that must be addressed. In this dissertation, we focus on the result integrity of outsourced data mining computations, one of the most important security concerns of the DMaS paradigm.;We propose efficient and practical approaches to verify the integrity of the data mining results that are returned by a potentially untrusted DMaS service provider (server). We propose both deterministic and probabilistic approaches that provide different degree of integrity guarantee. The deterministic approaches verify the integrity of the outsourced data mining results with 100% certainty, while the probabilistic approaches provide high probabilistic integrity guarantee with small overhead. For all the approaches, we provide both theoretical analysis and empirical study to show their effectiveness and efficiency.;We also study the problem of integrity verification of outsourced privacy preserving data mining computations. We focus on the randomization-based privacy preserving data mining technique that perturbs the dataset randomly to protect data privacy. We design efficient, robust result integrity verification approaches for such privacy-preserving data mining computations. Our empirical study illustrates the robustness and efficiency of our approaches.
展开▼
