LAUSR

Privacy-preserving data mining is an embryonic research area that addresses the integration of privacy-preserving concerns to data mining techniques. Classification is a problem in data mining which builds a model to classify the data and then identify the class label of unknown data based on the constructed model. Large amount of data is necessary to build a more accurate classifier. Sharing of data is one of the solutions to have enormous amount of data. When sharing the data among business associates, some sensitive patterns which can be derived from the data need not be revealed to the others. This situation raises a motivating issue of retaining the shared data with high quality by hiding some sensitive patterns. This paper addresses the problem of classification rule hiding by projecting a novel method based on data distortion approach. To select the best possible way of altering the instances and then selecting the optimal instances which reduces the loss of non-sensitive classification rules, a computational intelligence technique binary firefly algorithm is adapted with necessary changes. The transformed data set will be shared to the others which reveals only non-sensitive knowledge. A set of experiments were carried out to estimate the effectiveness of the proposed approach against existing similar ones by considering the performance measures miss cost, artifacts and deviation between original and transformed data sets. The experiments and comparisons have proved that the projected method preserves the privacy of sensitive classification rules as well as maintains quality of the transformed data set also.