Privacy Preserving Classification By Using Modified C4.5

Protecting the datasets supplied to third parties for data mining purposes is essential so that these datasets cannot be used for secondary purposes. C4.5 is a classification algorithm which works on mixed datasets. Data perturbation is an important technique in data privacy. This paper proposes a modified C4.5 which uses perturbed and unrealized datasets for classification. The decision tree is built by using the gain ratio as the split criteria and it is computed using the unreal and perturbed datasets. Experimental results are obtained by simulation in Weka.