A Decision Tree Based on Differential Privacy
Daozhu Sun, Nan Li, Shudan Yang, Qiming Du · 2021 IEEE 5th Information Technology,Networking,Electronic and Automation Control Conference (ITNEC) · 2021
In the era of big data, personal privacy leakage has become a hidden danger for social development. As an output privacy protection model, differential privacy can effectively quantify and limit personal privacy disclosure. Combining differential privacy with decision tree algorithm, a decision tree algorithm is proposed to meet differential privacy protection. The algorithm uses exponential mechanism to select split nodes and Laplace mechanism to add noise to leaf nodes. When allocating the privacy budget, the correlation and quantity of the data set are fully considered, which saves the privacy budget and realizes the dynamic change of the privacy budget with the change of the size of the data set. The experimental result on Adult data set display algorithm effectively improves the accuracy of classification result prediction under the premise of privacy protection.