HRL-DeepNet: A Hybrid Residual Layer Deep Neural Network for Cybersecurity Policy Modeling, Structuring, and Protecting Assets of Organizations

The widespread integration of interconnected network elements within the Internet of Things (IoT) has increased its vulnerability to security breaches. This is due to the various software and networks involved in IoT. Numerous elements within these networks lack built-in cyber defenses. Traditional methods like access control, password security, data authentication, malware scanners, and firewalls often fail against sophisticated cyber-attacks due to their reactive nature and limited adaptability. Additionally, intrusion detection systems and security audits can be prone to attacks and may struggle with evolving threats. To address these limitations, We propose a novel hybrid residual layer deep neural network (HRL-DeepNet) for detecting cyber-attacks and anomalies in organizational assets. The HRL-DeepNet employs gated recurrent unit (GRU), bidirectional long short-term memory (BiLSTM), and long short-term memory (LSTM) sequences in hybrid and residual setups. Utilization of hybrid and residual setups not only boosts the distinctiveness of the features but also improves the accuracy of intrusion detection. The proposed HRL-DeepNet, when evaluated on ToN-IoT and CICIDS2017, resulted in high accuracy, with a significantly low false positive rate (FPR) outperforming other state-of-the-art frameworks. Furthermore, the proposed HRL-DeepNet achieves accuracies of 0.999 and 0.986 on the ToN-IoT and CICIDS2017 datasets, respectively, while also achieving F1 scores of 0.977 and 0.966 on the same datasets. This demonstrates its superiority over recently reported works.