基于匿名化样本ID的安全多方集成树模型

s a widely used machine learning algorithm with strong interpretability and high adaptability, the ensemble decision tree models have been used in various fields, such as finance and medicine. However, the data in these fields is often strictly private and must not be disclosed and shared freely. Therefore, how to use privacy-related technologies to securely and efficiently perform joint modelling of ensemble tree models has become a research hotspot in industry and academia. At present, the private information of the sample space contains in the nodes in the ensemble decision tree models is generally expressed on the basis of the secret shared bit vector, but this method will bring a large number of redundant zeros and increase the computational cost; in addition, the calculation process of the information gain function involves operation inefficient secure division operator in existing methods.To solve the above problems, this paper proposes a decision tree node representation method based on anonymised sample ID. In this paper, aiming at the redundant zeros problem of node sample space representation method, the anonymized sample ID based on the exchangeable encryption system is used to represent the sample space of the node; for the inefficient safe division operator in the calculation process of information gain function, which can be transformed into efficient safe addition and multiplication operators by common division and simplification. Experiments on public datasets demonstrate that the proposed model has improved computational efficiency and has almost no loss compared with existing centralised training models.