BBBC-DDRL: A hybrid big-bang big-crunch optimization and deliberated deep reinforced learning mechanisms for cyber-attack detection

Abstract

Currently, the wireless networks are increasingly used in all kinds of applications due to their adaptability, flexibility, and easy-access. Without a predetermined architecture, wireless sensor networks are often deployed and unmanaged. Due to these traits, the wireless networks are more vulnerable to attacks, and an opponent can easily track traffic because the data are transmitted over the air. In the existing studies, a different types of attack detection methodologies are used for intrusion identification and classification. But, they having problems in terms of inaccurate prediction, high false positives, and lack of reliability. Thus, the proposed research work intends to implement a sophisticated and novel cyber-security model for safeguarding wireless networks. Here, the Graph Referencing Normalization (GRN) algorithm is specifically used to preprocess the datasets based on the normalization vector. Then, the cutting-edge Big-Bang Big-Crunch (BBBC) optimization technique is employed to obtain the relevant features from the normalized data for analyzing the characteristics of attacks. The Deliberate Deep Reinforced Learning (DDRL) classification process is used to predict the normal and attacking data based on the features. By using the GRN, BBBC and DDRL mechanisms, the intrusions are accurately predicted from the cyber-datasets. To assess the performance of this model, the Matlab simulation tool has been used, where the results are validated with the use of different parameters like precision, recall, f1-score, and etc. The estimated results indicate the average attack detection rate is improved up to 99% for all datasets, when compared to the other existing security models.