Adaptive Differential Privacy Budget Allocation Algorithm Based on Random Forest

Abstract

AbstractIn the context of big data, privacy protection in the process of data mining has become a hot issue in the field of security. The commonly used privacy protection method is to add differential privacy in the process of data mining, but the unreasonable privacy budget allocation leads to unsatisfactory classification results. To solve this problem, an adaptive differential privacy budget allocation algorithm RFDPP-weight based on random forest is proposed. Construct multiple decision trees with balanced features, calculate the Balance Error Rate (BER) of the data outside the bag in order to calculate the feature weight and then reconstruct the feature set. And the random forest with higher classification performance was constructed based on the new feature set. The decision tree weight was calculated according to the feature weight, and the privacy protection budget was adaptively allocated according to the decision tree weight. In the final prediction stage, the classification result with the largest weight of the corresponding decision tree is taken as the result of the random forest algorithm, so as to further improve the classification accuracy of the random forest algorithm. The experimental results show that the F1 score of our algorithm reaches 0.977 and 0.893 with a small number of decision trees in the Mushroom and Adult data sets, which proves that the algorithm can reasonably allocate the privacy budget and further improve the availability of data while protecting data privacy. KeywordsBig dataData miningPrivacy protectionDifferential privacyRandom forestBalance error rate