Modified Privacy Preserving Data Mining System for Improved Performance

Privacy of information and security issues now-a-days has become the requisite because of big data. A novel framework for extracting and deriving information when the data is distributed amongst the multiple parties is presented by Privacy Preserving Data Mining (PPDM). The concern of PPDM system is to protect the disclosure of information and its misuse. Major issue with PPDM that exists is to use the coherent data mining algorithm for preserving privacy of data. Various PPDM techniques have been proposed till now. One of them is DPQR (Data Perturbation & Query Restriction) algorithm, which is implemented only on Boolean data. In proposed approach, the SVD (Singular value decomposition) data perturbation technique is applied for data modification; discretization of raw data is done and generates the perturbed/distorted data. SVD technique improves the level of privacy protection by providing higher degree of data distortion. The algorithm is applied on perturbed data with association rule mining and Hamiltonian matrix concepts to find out frequent itemsets. By this the confidential data is preserved. The performance metrics for the approach are privacy level, efficiency, scalability and data quality. The time required for calculating matrix inversion is reduced by using Hamiltonian matrix. Main performance metrics is the privacy level and the privacy preserving degree is improved by dimensional reduction based perturbation i.e. multi dimensional perturbation (SVD and NMF) data perturbation techniques. The novelty of the proposed method is, it is being applied on numeric data and expected to achieve comparable parameters as shown on Boolean data.