A Differential Privacy Budget Allocation Algorithm Based on Out-of-Bag Estimation in Random Forest

The issue of how to improve the usability data publishing under differential privacy has become one top questions in field machine learning protection, and key solving this problem is allocate a reasonable protection budget. To solve problem, we design budget allocation algorithm based on out-of-bag estimation random forest. firstly calculates decision tree weights feature by protection. Secondly, statistical methods are introduced classify features into best set, pruned removable set. Then, pruning performed using set avoid trees over-fitting when constructing an ?-differential Finally, allocated proportionally We conducted experimental comparisons with real sets from Adult Mushroom demonstrate that not only protects security privacy, but also improves model classification accuracy availability.