结合特征选择与集成学习的密码体制识别方案

在密文识别过程中,加密算法是进一步分析密文的必要前提。然而现有密文识别方案存在形式单一的问题,并且在识别多种密码体制时难以应对不同密码体制间存在的差异。分析密文特征对识别效果的影响机制,结合Relief特征选择算法和异质集成学习算法,提出一种可适应多种密码体制识别情景的动态特征识别方案。在36种加密算法产生的密文数据集上进行实验,结果表明,与基于随机森林的密码体制分层识别方案相比,该方案在3类不同密码体制识别情景下的识别准确率分别提高了6.41%、10.03%和11.40%。

Cryptosystem Identification Scheme Combining Feature Selection and Ensemble Learning

In cyphertext identification,the encryption algorithm is the prerequisite for further analysis of ciphertext.The existing identification schemes are constructed in a single form,and thus often fail to cope with the differences between different cryptosystems when identifying multiple cryptosystems.To address the problem,this paper studies how different ciphertext features influence the performance of identification schemes,then combines the Relief feature selection algorithm and heterogeneous ensemble learning to propose a dynamic feature identification scheme that can adapt to the scenario of multiple cryptosystem identification.Experiments are carried out on ciphertext data sets generated by thirty-six encryption algorithms,and results show that,compared with the existing hierarchical cryptosystem identification schemes based on random forest,the proposed scheme increases the identification accuracy by 6.41%,10.03%and 11.40%respectively in three different cryptosystem identification scenarios.