Research on the Method of Selecting the Optimal Feature Subset in Big Data for Energy Analysis Attack

At present, the application of machine learning in energy analysis attack is a research hot spot of energy analysis attack, and the selection of feature points is an important factor that affects the machine learning model, how to choose the optimal feature subset is a key factor related to the success or failure of energy analysis attack. AES algorithm emphasizes to increase the complexity of encrypted data by a large number of encryption rounds. It generally runs ten rounds of encryption operation, but the energy information studied by attackers is only a part of one round in ten rounds. Therefore, it is of great significance to effectively select the optimal feature subset with the least amount of data from a large number of data for energy analysis attacks. According to the characteristics of high-dimensional small features of energy data, this paper proposes a new optimal feature subset selection method-secondary feature selection method named F-RFECV. Firstly, the F-test is used to quickly eliminate a large number of irrelevant and redundant features to initially select candidate energy feature subsets, and then the redundant features are further eliminated by recursive feature elimination and cross validation, so as to obtain the optimal energy feature subset, which effectively realizes the problem of small feature recognition in high-dimensional features, thus improving the success rate of model attack in subsequent machine learning. Experiments show that the attack success rate can be increased by 17% by using the secondary feature selection method (F-RFECV).