Differentially Private Linear Regression Analysis Based On Laplace Mechanism

With the advent of the era of big data, data analysis occurs, statistics and publishing applications, making the relevant social institutions can get a lot of personal data and corporate data mining and analysis, leading to the commercial value and scientific value. Major shopping malls such as customer data, customer information all real estate purchase, the user’s mobile business hall business process information, customer information, banks, securities companies and individual transactions and statistical analysis. However, these data relate to a lot of private information, once released and analysis of these data, are faced with the problem of loss of privacy, how to protect data privacy in order to prevent disclosure of sensitive information has become a daunting task currently facing.Differential current data privacy protection technology is the most important release of privacy protection methods. It does this by adding noise to the query data to interfere attacker leaked raw data object, so as to achieve privacy protection. Application of differential privacy technology makes data publishing efficiency has been greatly improved, but the difference in order to meet privacy requirements need to inject excessive noise, affect the accuracy and reliability of data, resulting in low-quality results.Linear regression analysis focuses on privacy Laplace differential mechanism herein, the method of regression analysis of the results of the objective function instead of a differential noise interference of privacy, a regression analysis on each data set, with the remaining properties to predict annual income, the final error rate compare and analyze the results. Experiments show that the method proposed in this paper not only protect the user’s private data, do not affect the validity of linear regression analysis.Thesis done the following work:(1) Application and Research Office deficiencies differential current privacy model fitting part. (2) Based on theoretical analysis and differential privacy model works. (3) the use of privacy Laplace differential mechanism constructed linear regression model, and the model optimization and evaluation.