Machine Learning Based Attack on Certain Encryption Schemes

Abstract

Machine learning provides a very promising approach to attack cryptographic implementations. In this paper, a machine learning based attack on text encrypted with public key encryption schemes like RSA and Elliptic Curve Cryptography (ECC) is shown. Decision Trees, a popular approach to perform classification are utilized as multi-class classifiers to learn the structure of the text from training examples so that an unknown similar text can be decrypted successfully. The attack is performed on a section of the Enron email dataset. Three different feature sets are created and then their individual performance is evaluated. Finally, their results are combined together to find out the total percentage of correct partial decrypted text in the test set.