Virtual Network Intrusion Detection Using Random Forest Salp Swarm Optimization-based Feature Selection and a Deep Sparse Auto Encoders

Abstract

With the increasing use of information technology, hacking has become more prevalent than ever before in every domain of life. Providing privacy and protection is a challenging task for the developers of security management systems. Current intrusion detection systems (IDSs) have the disadvantages of high running time, high false positive rate, and low detection accuracy. In this study, a hybrid feature selection algorithm was introduced. A deep sparse auto encoder (DSAE) method based on deep learning was proposed for intrusion detection. The proposed hybrid approach consisted of random forest (RF) and salp swarm optimization (SSO). Five modules were used in this IDS. Data were collected from the data set. Feature selection and classification modules are the most crucial modules in IDSs. The hybrid deep learning approach was used for optimal feature selection and classification. Alert module demonstrate results based on the decision manager, and information and rules are stored in the knowledge-based module. The proposed model was implemented on the Python platform. Performance metrics such as the detection rate, false acceptance rate, accuracy, Kappa statistics, and CCI of the proposed model (RFSSO–DSAE) were compared with those of other existing methods such as RF gravitational search algorithm–SAE (RFGSA–SAE), RF genetic algorithm–SAE (RFGA–SAE), RFSSO–artificial neural network (RFSSO–ANN), RFGSA–ANN, RFGA–ANN, RFSSO–support vector machine (RFSSO–SVM), RFGSA–SVM, and RFGA–SVM. The proposed model exhibited superior performance.