Master’s Thesis Defense at the College of Information Technology on Anomaly Detection in Networks Using a Hybrid Deep Learning Approach

Duhaa Fadill Abbas
The Department of Information Networks at the College of Information Technology – University of Babylon hosted the defense of the Master’s thesis submitted by the student Fatima Haider Jasim, entitled: “A Hybrid Deep Learning Approach for Network Anomaly Detection Based on Sparse Autoencoder and Convolutional Neural Network.” The thesis was supervised by Assistant Professor Dr. Amir Kazem Hadi. The defense was conducted in a formal academic setting that reflected the advanced level of the research and the growing scholarly interest in network security and modern artificial intelligence techniques.

The thesis highlighted the rapid and continuous evolution of cyberattacks and the significant challenges they pose to traditional signature-based intrusion detection systems, which often suffer from limited capability in identifying novel attacks and weak generalization performance when dealing with underrepresented attack classes.

The researcher proposed an innovative hybrid system based on deep learning techniques, integrating a Sparse Autoencoder (SAE) for high-quality feature extraction and dimensionality reduction with a Convolutional Neural Network (CNN) to achieve accurate classification of network traffic and intrusion detection. Furthermore, an attention mechanism was incorporated into the system architecture to enhance the model’s ability to prioritize the most informative and contextually relevant features in identifying anomalous behavior.

The experimental results demonstrated the superiority of the proposed system when evaluated on benchmark datasets such as NSL-KDD and CICIDS2017, achieving high detection accuracy compared to existing studies, with a reported accuracy of 99% on the CICIDS2017 dataset. In addition, the system achieved 100% accuracy when tested on a real-world dataset, indicating its strong performance in practical deployment scenarios.
To further assess real-time applicability, the researcher conducted an online simulation experiment that emulated actual network traffic conditions. The findings revealed the system’s capability to perform dynamic, real-time anomaly detection and to respond rapidly to potential threats.
The study concluded that the proposed approach represents a significant contribution to the field of network security by combining efficient feature extraction with high-precision classification, thereby enhancing the reliability and robustness of intrusion detection systems and addressing the escalating challenges in modern network environments.