Robust Cyber Defense Framework with Adamax Convolutional Neural Network and Hierarchical Multi-Scale Long Short-Term Memory

Abstract

Cybersecurity protects networks, systems, and data from theft, loss, or unauthorized access. The fast emergence of sophisticated cyber threats like ransomware, phishing, and advanced persistent threats, which consistently surpass established defense measures, is one of cybersecurity’s many difficulties. The increasing number of Internet-of-Things devices and the growing complication of IT environments offer enormous attack surfaces that are challenging to secure and monitor. The lack of qualified cybersecurity specialists further exacerbates the issue, making it difficult for enterprises to implement efficient security measures. To overcome these limitations, this research proposes a novel Adamax-optimization convolutional neural network with hierarchical multi-scale long short-term memory (AOCNN-HMLSTM) to detect cyberattacks. The model incorporates multiple crucial phases: preprocessing, feature mapping, temporal feature learning, and classification. The model uses an AOCNN to map features, an HMLSTM to capture temporal features, and min-max normalization to first normalize the Car-Hacking dataset. The obtained features are then categorized into normal operation, gear spoofing, RPM spoofing, denial of service, and fuzzy attacks employing a fully connected layer and softmax layer for multi-class classification. As a result, the performance metrics provide significant implementation results, such that accuracy (98.8%), precision (99%), and recall (98%) attain the finest results compared with several existing studies.