Product data protection method based on differential privacy logistic regression

Abstract

The invention relates to a product data protection method based on differential privacy logistic regression, and aims to solve the problem of product data privacy protection in supply chain management. The traditional differential privacy technology has the risk of privacy disclosure when processing related data, so that the invention provides a method for combining related differential privacy and logistic regression to improve the effectiveness of product data privacy protection. According to the technical scheme, the method comprises the steps of data preprocessing, feature selection, correlation measurement, noise addition, model training and prediction and the like. A most relevant feature subset is selected by using a forward-backward feature selection method. A correlation difference privacy mechanism is introduced into a logistic regression model, correlation between data records is considered, and data privacy is protected by adding Laplace noise or Gaussian noise. Experimental results show that the privacy protection level of product data can be effectively improved, and meanwhile high prediction accuracy is kept. The method has the beneficial effects that through the combination of a related differential privacy mechanism and a feature selection method, the data privacy protection capability is improved, the data query error is reduced, and the practicability of data and the accuracy of a prediction model are improved.

Description

Technical Field

[0001] The present invention relates to data privacy protection technology, and in particular to a product data protection method applied in supply chain management, specifically a data protection method combining a relevant differential privacy mechanism and a logistic regression model. Background Art

[0002] In supply chain management, the security and privacy of product data are crucial. These data include product design parameters, manufacturing process data, material specifications, historical sales data, etc., which usually involve sensitive information and need to be protected when performing data analysis and prediction. Traditional differential privacy technology has the risk of privacy leakage when processing related data. Therefore, it is necessary to develop a new data protection method to improve the privacy protection level of product data. Summary of the invention

[0003] The purpose of the present invention is to provide a method based on differential privacy logistic regression for effectively protecting product data in the supply chain while ensuring the accuracy of data processing. The present invention achieves the above purpose through the following technical solutions:

[0004] (1) Data preprocessing step: cleaning, integrating and converting product data;

[0005] (2) Feature selection step, selecting the optimal feature subset through the forward-backward feature selection method;

[0006] (3) The correlation measurement step constructs the data correlation matrix and adds noise according to the correlation sensitivity;

[0007] (4) Noise addition step, adding noise during the training process of the logistic regression model;

[0008] (5) Model training and prediction step: use the selected feature subset to train the model and predict product demand. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Attached Figure 1 : Product Data Protection Flowchart

[0010] Attached Figure 1 The overall process of product data protection is demonstrated, including steps such as data preprocessing, feature selection, association measurement, noise addition, and model training and prediction.

[0011] Attached Figure 2 :Association measurement diagram

[0012] Attached Figure 2 The working principle of the association measurement mechanism is demonstrated, including the construction of the data correlation matrix and the calculation of the correlation sensitivity.

[0013] Attached Figure 3 : Flowchart of noise addition and training prediction of logistic regression model

[0014] Attached Figure 3 The paper shows the steps of introducing noise into the training process of the logistic regression model to protect data privacy, and then the complete process of using this data for model training and prediction. DETAILED DESCRIPTION

[0015] 1. Data preprocessing:

[0016] 1) Collect product data, including product design parameters, manufacturing process data, material specifications, historical sales data, etc.

[0017] 2) Clean the data to remove erroneous values, missing values, duplicate values ​​and outliers.

[0018] 3) Integrate data and merge data from different sources to ensure data consistency.

[0019] 4) Perform unique-hot encoding on product information and normalize other data.

[0020] 2. Feature selection method:

[0021] 1) Use the random forest algorithm to calculate the importance of each feature in the dataset.

[0022] 2) Remove each feature and calculate its impact on the relevance of the dataset.

[0023] 3) Perform forward and backward traversal based on feature importance and data relevance to select the optimal feature subset.

[0024] 3. Association measurement:

[0025] 1) Construct a data correlation matrix and use the Pearson correlation coefficient and Mahalanobis distance to measure the correlation between records.

[0026] 2) Define correlation sensitivity to adjust the addition of noise based on the correlation between data records.

[0027] 4. Noise addition:

[0028] During the training of the logistic regression model, Laplace noise or Gaussian noise is added to the mean of the sample gradient to meet the privacy protection requirements.

[0029] 5. Model training and prediction:

[0030] 1) Train a logistic regression model using the selected feature subset.

[0031] 2) During model training, noise is added according to the relevant sensitivity to protect data privacy.

[0032] 3) Use the trained model to predict product demand.

Claims

1. A product data protection method based on differential privacy logistic regression, used to protect product data in supply chain management, characterized in that: The following steps are involved: (1) Data preprocessing step: cleaning, integrating and converting product data; (2) Feature selection step, selecting the optimal feature subset through the forward-backward feature selection method; (3) Correlation calculation step, constructing the data correlation matrix and adding noise according to the relevant sensitivity; (4) Noise addition step, adding noise during the training process of the logistic regression model; (5) Model training and prediction step: use the selected feature subset to train the model and predict product demand.

2. The method according to claim 1, characterized in that: The feature selection step includes calculating feature importance and data relevance using a random forest algorithm.

3. The method according to claim 1, characterized in that The association measurement step includes using Pearson correlation coefficient and Mahalanobis distance to measure the correlation between records.

4. The method according to claim 1, characterized in that: The noise adding step includes adding Laplace noise or Gaussian noise to the mean of the sample gradient.

Images