Privacy Preserving Decision Tree Prediction Scheme Based On Secure Multiparty Computation

With the rapid development of big data,cloud computing,Internet of things and artificial intelligence technology,how to effectively use data under the condition of protecting data privacy has become an urgent problem to be solved.In the current machine learning prediction service,the service provider has a trained machine learning model,anyone can submit their own data to obtain the corresponding prediction results.Machine learning prediction service makes people better enjoy the convenience of artificial intelligence to life,but customers need to submit the data directly to the service provider in plaintext,which also faces the problem of data privacy leakage.Therefore,privacy preserving machine learning prediction based on cryptography has a wide range of application scenarios.Decision tree algorithm is one of the most commonly used machine learning algorithms because of its simple structure and fast running speed.As an important technology in cryptography,secure multi-party computation has unique advantages in solving the problem of data privacy leakage,and is widely used in various privacy protection scenarios.Therefore,it is of great significance to construct a privacy preserving decision tree prediction scheme based on secure multi-party computation technology.This paper takes decision tree algorithm as the research object,and studies the privacy protection decision tree prediction scheme from the aspects of performance and security.Firstly,we analyze the current privacy preserving decision tree prediction scheme,and it is found that most of the schemes are constructed by using obfuscation circuit or homomorphic encryption.In this scheme,we use lightweight cryptography techniques such as secret sharing to construct decision tree prediction scheme,and avoid using complex cryptographic primitives such as obfuscation circuit and homomorphic encryption to improve the protocol performance efficiency.Secondly,our scheme makes use of the secure multi-party computation framework based on cloud server,improves the basic protocols of the secure three-party computation model,such as secret sharing,multiplication protocol,vector dot product protocol,bit extraction protocol,and applies them to the privacy preserving decision tree prediction algorithm,designs a new protocol by modules,and uses a lot of precomputation in the decision tree prediction process to reduces the complexity of computation and communication in the online phase.Moreover,different from the traditional predictor client model,the proposed model only requires the prediction service provider and the client to upload their models or data to the server in the form of secret sharing,and there is no need to maintain online communication to participate in the prediction process.Finally,the privacy preserving decision tree prediction scheme is divided into input preparation module,oblivious feature selection module,oblivious decision node comparison module,decision path generation module and prediction result generation module.The performance and security of these five modules are analyzed respectively.The feasibility of the proposed scheme is proved by ideal reality paradigm and module sequence combination theorem in order to ensure the data privacy of users,the privacy of intermediate calculation results,the privacy of decision tree model and the privacy of prediction results under the model of semi honest adversary.