Differential Privacy Protection Algorithm And Its Financial Application

In the era of the digital economy,the rapid development of data sharing has also made people face the problem of privacy leakage.With the introduction of the EU's privacy protection regulation GDPR,more and more enterprises and individuals have begun to pay attention to the research and application of privacy protection.However,since it is difficult to have specific measurement standards for privacy,and most privacy protection methods require a large amount of computational costs,it is difficult to apply them to real life.Differential privacy,as a privacy protection method that can quantify privacy,have relatively little computing overhead and energy overhead,and has flexible combination characteristics,is widely used by major data-intensive companies in practice.This paper proposes a differential privacy protection algorithm for decision trees and random forests,and conducts empirical analysis with data sets in the financial field.Firstly,the related theories involved in decision trees and differential privacy algorithms are introduced,and then classification algorithms such as decision trees and random forests for differential privacy protection are derived.The algorithm uses the Laplace mechanism to handle discrete features,the Exponential mechanism to handle continuous features,and then selects the best split feature and split point,and adds noise to the count value of the leaf nodes.In order to verify the usability of the designed algorithm,the paper uses credit card data set and household financial data set,and selects classification accuracy as the evaluation index,and conducts empirical analysis on the designed algorithm under different parameter combinations.Finally,this paper summarizes the results of the paperand,and gives some advice on future research directions and applications from the aspects of the application and demand of differential privacy in practice,and the optimization of the privacy protection budget allocation mechanism.The algorithm proposed in this paper uses an equal budget allocation mechanism to allocate privacy budgets.While making full use of privacy protection budgets,the accuracy of the model is improved.This strategy of selecting the optimal features saves the time cost of the algorithm,reduces the added noise,and improves the performance.Algorithm performance.From the experiment,we can see vb that the Diff P-CART algorithm and Diff PRF algorithm proposed in the paper can have higher classification accuracy while protecting data privacy and security.