Method for constructing decision tree based on differential privacy protection

Abstract

The invention relates to a method for constructing a decision tree based on differential privacy protection. The method comprises the steps that sampling is performed on an original dataset according to a sampling probability p to obtain dataset samples, and the obtained dataset meets the requirement for ln(1+p(e<epsilon>-1)-differential privacy, wherein primary processing is performed on the dataset obtained through sampling, and continuous properties and discrete properties are made to jointly participate in decision selection under privacy protection; a C4.5 decision tree is initialized according to extracted dataset samples, and a sparse vector method is utilized to judge whether nodes in the decision tree continue to split; and the decision tree is constructed recursively. Through the method, classification accuracy is high, and the decision tree can be constructed efficiently and accurately while privacy is protected.

Description

technical field [0001] The invention relates to a decision tree privacy protection method, in particular to a method based on differential privacy protection decision tree. Background technique [0002] With the development of hardware and technology, it is not a difficult problem to collect a large amount of data in a timely and effective manner, but how to mine useful knowledge and value from these data is a difficult point for people to study. Classification algorithm is a commonly used data mining tool. It can well support applications such as precise marketing, personality preference and credit analysis, and is widely loved by the financial industry and companies. Decision tree is one of the common classification algorithms. When building a decision tree, you first need to decide which attribute to split the node on. This decision is dominated by the data in the node. In addition, once the decision tree is constructed, the leaf nodes can output count information about ...

AI Technical Summary

Problems solved by technology

However, they mainly have two shortcomings: 1) decision classification is only performed on small spatial data. When the data points reach millions of levels, a large number of classification trees will be generated, resulting in inefficient selection methods; 2) when building a decision tree In the process, the privacy cost will be allocated layer by layer. When the height of the tree is very large, the privacy budget may be exhausted

Images