Prognosis by means of privacy preserving data mining in a partitioned distributed database for credit card in banking

Abstract

quot newline newlineBackground: newlineData mining in simple words is the extraction of unseen informative patterns from huge amount of data. When data mining is carried out with data from multiple entities, the protection of delicate personal information always remains an apprehension. A wide variety of solutions have been suggested and implemented in this area, which are domain, problem specific, each having limitations and advantages. The problem addressed is of privacy preserving data mining for credit card applicants. It is observed that this situation is conceded with human judgements and their credit score. newline newlineAim: newlineThe task undertaken is to mine the data sets from different locations with an essence such that all over data remains maximum concealed but still accurate mining results are available to everyone. The joint model developed is used for classification, aims at providing more precision in forecasting. newline newline newlineMaterials and Methods: newlineThe experiment has been carried in a virtual environment with five distinct data sources and one middle party source. The data sets under consideration are from a sensitive domain of credit card applicants of multiple banks and hence real life figures were not possible to obtain. newlineThe use of synthetic data has been made, which happens for most of the experiments under this theme. newline newlineResults and Discussion: newlineThe resultant classifier is functional for determining whether the applicant should be approved of card or not. Also, the subsequent combined models give better results of accuracy as compared to those with only individual replica. The experiment has been done in two different ways with two different arrangements of the model. With massive data sets, the second model may perform marginally faster as compared to the first model. newline newlineConclusion(s): newline With the experimentation, the following inference can be drawn. newline(a) The joint models along with desired level of privacy preserving is high on robustness and efficiency, which could turn out to be great help in dropping the rate of the condition when the