A differentially private random decision forest using reliable signal-to-noise ratios

When dealing with personal data, it is important for data miners to have algorithms available for discovering trends and patterns in the data without exposing people’s private information. Differential privacy offers an enforceable definition of privacy that can provide each individual in a dataset a guarantee that their personal information is no more at risk than it would be if their data was not in the dataset at all. By using mechanisms that achieve differential privacy, we propose a decision forest algorithm that uses the theory of Signal-to-Noise Ratios to automatically tune the algorithm’s parameters, and to make sure that any differentially private noise added to the results does not outweigh the true results. Our experiments demonstrate that our differentially private algorithm can achieve high prediction accuracy.