A differential approach and deep neural network based data privacy-preserving model in cloud environment

Abstract

Data outsourcing has become indispensable to allow information sharing among multiple parties. The users do not fully trust the cloud platform since it is operated by a third party. Preserving privacy while sharing the data among different parties is a challenging task; therefore, users apply the differential privacy mechanism to protect their data. However, such protection mechanisms suffer from the problem of degradation of learning results. In this paper, the authors address the degradation of the learning results due to noise injection into user’s data through $\epsilon$-differential privacy. A differential approach and deep neural network based data privacy-preserving model is proposed, which injects noise at an appropriate position by exploiting the properties of the Laplace transform to maintain the accuracy level. The experiments are conducted over Steel Plates Fault, Spambase, Banknote Authentication, and Monk Problem datasets for deep neural network classifier to evaluate the model’s efficiency in accuracy, precision, recall, and F1-score terms. The achieved results show that the proposed model ensures high accuracy, precision, recall, and F1-score up to 99.75%, 99.72%, 99.72%, and 99.72%, and improvement up to 19.34%, 30.67%, 29.39%, and 32.11%, respectively, as compared to existing approaches.