Efficient privacy-preserving decision tree classification protocol

Abstract

To provide privacy-preserving decision tree classification services in the Internet of things (IoT) big data scenario, an efficient privacy-preserving decision tree classification protocol was proposed by adopting the secure multiparty computation framework into the classification model.The entire protocol consisted of three parts: the original decision tree model mixing, the Boolean share-based privacy-preserving comparing, and the 1-out-of-n oblivious transfer-based classification result obtaining.Via the proposed protocol, the service providers could protect the parameters of their decision tree models and the users were able to derive the classification result without exposing their privately hold data.Through a concrete security analysis, the proposed protocol was proved to be secure against semi-honest adversaries.By implementing the proposed protocol on various practical decision tree models from open datasets, the classification accuracy and the average time cost for completing one privacy-preserving classification service were evaluated.After compared with existing related works, the performance superiority of the proposed protocol is demonstrated.

Keywords

• decision tree
• privacy preserving
• oblivious transfer
• secure multiparty computation