An Optimized Deep Learning Approach for Intrusion Detection: AE-DBN Hybrid Model with Dingo Feature Selection on CSE-CIC-IDS2018
Srimaan Yarram · 2025
The Internet of Things (IoT) enables interconnected, always-on smart devices. Attacks on network communications made possible by these interconnected devices provide a problem for cybersecurity systems. Threats to system and user operations from these types of assaults have persisted. So, one of the most popular ways to keep these vulnerabilities protected from cyberattacks is with Intrusion Detection Systems (IDS). Traditional intrusion detection systems are made more difficult by the ever-changing and multi-faceted nature of the threats that affect IoT networks. The need for more sophisticated intrusion detection systems (IDS) to detect and prevent cyber-attacks has grown in response to the sophistication of network security threats. Our paper presents a deep learning model for network attack detection that combines Autoencoder (AE) and Deep Belief Network (DBN), together with an optimised feature selection method based on the Dingo Optimiser (DO). The CSE-CIC-IDS2018 dataset, an extensive dataset for intrusion detection research, is used to test the model. In this method, DBN is used for its strong classification skills, and AE is used for efficient feature extraction and dimensionality reduction. To reduce computational complexity and improve model performance, the Dingo Optimiser is used to choose the most relevant features. The experimental results show that our strategy greatly improves the accuracy of attack detection while reducing the rates of false alarms. Accuracy, precision, recall, F1-score, and detection rate are all areas where our hybrid method outperforms traditional ML and DL models. This study's results show that intrusion detection systems that use a mix of deep learning classification, unsupervised feature learning, and metaheuristic optimization perform better. Cybersecurity applications can benefit from this model's potential for real-time deployment and resistance to adversarial attacks, and it offers a viable approach to enhancing network security.