Prognosis using distributed data classification with privacy preserving: A novel approach

Privacy preserving data mining (PPDM) is a captivating forte for every researcher who has been closely pursuing data mining, for its inherent nature of ubiquitous pervasiveness. As few years back, data mining was essential and vital to any sphere, so is the now the spectrum of privacy preserving data mining expanding with a thrust upon its applicability and efficacy. PPDM is a pool of solutions which takes care of shielding of data which has personal or private information and where any level of percolation of such information can be a cause of colossal and irreversible loss to an individual or business. At the same time, PPDM is also concerned with not compromising on the utility of other data which would be participating in mining. A balance between both the aspects: the secrecy and accuracy requires a smart balancing solution. Any algorithm suggested vary in several measures like efficiency, accuracy, data transfer costs, level of secrecy maintained, speed: to name a few. No algorithm is such that it can be generalized to perform superior to the rest. They are situation, domain and requirement specific. In this paper, an algorithm with a background framework for PPDM is proposed which anonymizes sensitive horizontal partitioned style distributed data, before they partake in collective mining process. Efforts have been made to conceal maximum personal information and not allowing it to affect on the results of mining. It is also kept in mind that the data transfer remains minimal during the entire process without distressing the quality of final findings. The experimental results and analysis is also presented for a detailed evaluation of the proposed method. An earlier solution in the same genre and environment is compared with the existing solution on important aspects.