Find in Library

Random forest is a simple and effective model for ensemble learning with wide potential applications. Implementation of random forest evaluations while preserving privacy for the source data is demanding but also challenging. In this paper, we propose a practical and fault-tolerant privacy-preserving random forest evaluation scheme based on asymmetric encryption. The user can use asymmetric encryption to encrypt the data outsourced to the cloud platform and specify who can access the final evaluation results. After receiving the encrypted inputs from the user, the cloud platform evaluates via a random forest model and outputs the aggregated results where only the designated recipient can decrypt them. Threat analyses prove that the proposed scheme achieves the desirable security properties, such as correctness, confidentiality and robustness. Moreover, efficiency analyses demonstrate that the scheme is practical for real-world applications.