Publication

Abstract : The blistering development of cyber threats due to more advanced and covert attack techniques has revealed the major shortcoming of the traditional intrusion detection mechanisms based on the fixed rules or one-dimensional analysis. Conventional methods tend to fail in the process of balancing detection and reduction of false-positive in dynamic networks. Inspired by these issues, this paper suggests a safe signal and image processing architecture, which relies on machine learning, next-generation cyber defense. The suggested framework uses a dual-branch architecture where the network traffic is processed as multivariate signals and image representation obtained after signal-to-image conversion. One branch of signal processing uses statistical feature learning and another branch of image processing uses convolutional neural networks to learn both spatial and nonlinear traffic patterns. Integration of complementary representations is then done at the feature-level to produce final intrusion classification. The framework is tested on the benchmark dataset of UNSW-NB15 in a supervised learning environment, and the performance is measured with the help of conventional measures such as accuracy, precision, recall, F1-score, and ROC-AUC. The experimental findings show that the suggested fusion architecture has better robustness, better discriminative ability and lower false-positive rates than single-modality models. These results emphasize the efficiency and useful applicability of secure signal image learning architectures in implementation in next generation cyber defense systems.