Secure Frequent Pattern Mining by Fully Homomorphic Encryption with Ciphertext Packing

Abstract

We propose an e cient and secure frequent pattern mining protocol with fully homomorphic encryption (FHE). Nowadays, secure outsourcing of mining tasks to the cloud with FHE is gaining attentions. However, FHE execution leads to significant time and space complex- ities. P3CC, the first proposed secure protocol with FHE for frequent pattern mining, has these particular problems. It generates ciphertexts for each component in item-transaction data matrix, and executes nu- merous operations over the encrypted components. To address this issue, we propose e cient frequent pattern mining with ciphertext packing. By adopting the packing method, our scheme will require fewer ciphertexts and associated operations than P3CC, thus reducing both encryption and calculation times. We have also optimized its implementation by reusing previously produced results so as not to repeat calculations. Our experimental evaluation shows that the proposed scheme runs 430 times faster than P3CC, and uses 94.7% less memory with 10,000 transactions data.