MINING DISTRIBUTED PRIVACY PRESERVED DATABASES USING EVOLUTIONARY ALGORITHMS

Abstract

Data mining has been defined as "The nontrivial extraction of implicit, previously unknown, and potentially useful information from data". Data mining includes tasks like frequent pattern mining, classification, clustering, association rule mining etc. These technique can be applied on different type of data like medical data, market basket data etc. The frequent pattern mining is used to fmd interesting frequent pattern from databases. Classification is the task of supervised learning i.e. the learning of the classifier is supervised in that it is told to which class each training tuple belongs, classification can be done in different ways like neural network, -probabilistic, evolutionary etc. To extract information, a database may be considered as a large search space, and a mining algorithm as a search strategy. In general, a search space consists of an enormous number of elements, which makes it infeasible to search exhaustively. As a search strategy, evolutionary -algorithms have been applied successfully in many fields. An evolutionary algorithm (EA) is a search technique used in computing to fmd exact or approximate solutions to optimization and search problems, due to their property of `robustness with respect to local maxima/minima' and `domain independent nature'. In many real world applications such as hospitals, retail-shops, design-firms and universities databases, data is distributed across different sources; the key goal for privacy preservation data mining is to allow computation of aggregate statistics over an entire data set without compromising the privacy of private data of the participating data sources. In our work, we have proposed the modified evolutionary rule-classifier, modified on its fitness function, attribute probability distribution and chromosome representation. Further we have proposed an evolutionary frequent pattern algorithm which is applied on a distributed scenario viz, horizontal, vertical and arbitrary partition. Then association rule mining is performed. A set of algorithm are proposed for preservation of privacy while finding frequent pattern. For privacy preservation we have used 'encryption and' also suggested a modified secure sum algorithm with two offset. The datasets chosen are nursery dataset and market basket dataset. The implementation of the proposed algorithms and all its prerequisites are coded using C++ programming language and C Socket Programming.