Performance Comparison of Security Level of Cryptosystems by using Machine Learning

Abstract

Recent developments in multimedia technologies have raised concerns about digital data security. However, many proposed encryption algorithms have been proven insecure over the past few decades, presenting a serious security risk to sensitive data. The best encryption technology should be used as protection against these attacks, but the type of data to be protected will determine the right algorithm in each case. Comparing different encryption systems one by one to find the best one can take a lot of time. We present a security level determination method for image encryption schemes that uses a support vector machine (SVM) for quickly and accurately selection of appropriate encryption algorithms. Furthermore, the dataset uses common cryptographic security techniques including entropy, contrast, homogeneity, peak-signal-to-noise ratio, mean square error, and energy. These variables are used as features for dividing the encryption algorithms extracted from the different images. Depending on the level of security, dataset labels are divided into three groups: weak, acceptable, and strong. The results show the advantage of Support Vector Machine (SVM) and we also improve the performance of the SVM by using XGBoost to improve the performance of our existing model.