Research on Privacy Protection Technology Based on Decision Tree Algorithm

In view of the deep contradiction between privacy protection needs and federated learning efficiency bottlenecks in the financial field, this paper proposes a federated decision tree model that integrates histogram data storage and multi-party collaborative prediction. By building a histogram splitting mechanism based on Gini index optimization, the coordinated optimization of feature selection and privacy protection in a distributed environment is achieved. In the model prediction stage, we break through the rigid constraints of "feature attribution rights binding computing execution rights" in traditional federated learning, and design local computing task pre-generation and cross-domain logic synthesis technology to effectively decouple the strong correlation between feature locations and computing paths. Experimental results show that this framework reconstructs data representation form and communication protocol, and forms a dynamic balance mechanism between computing resources and network bandwidth within the privacy and security boundary while maintaining the interpretability of the decision tree model. Experimental verification shows that on the basis of ensuring data sovereignty isolation, this method can better achieve the dual improvement of model effectiveness and privacy strength.