A K-nearest neighbor classifier based on homomorphic encryption scheme

Abstract

Homomorphic encryption technology can analyze the data stored in the cloud without decryption, because the results of ciphertext calculation after decryption are the same as the corresponding plaintext calculation results. Based on homomorphic encryption and machine learning technology, this paper proposes a K-nearest neighbor classifier based on homomorphic encryption scheme, Homomorphic encryption technology can not only ensure the security of the data, but also analyze the data in the ciphertext state since the characteristics of homomorphism, avoiding the data insecurity problem caused by analyzing the data after decryption in the clound. In this scheme, we first improve the ciphertext comparison algorithm and improve the judgment of sample label in ciphertext state. Then, using k-nearest neighbor classifier, a ring based selection algorithm is designed to reduce the time of ciphertext operation. The results show that our scheme can realizes the ciphertext classification On the condition of ensuring the accuracy of classification. Compared with the original k-nearest neighbor classification method, the classification accuracy of the our algorithm is improved about 1%, but the time cost is larger than the original k-nearest neighbor classification method.