PhD Dissertation at the College of Information Technology Explores Network Slicing and Efficient Resource Management in SDN for Traffic Classification Using an Optimized Deep Neural Network Algorithm
Duhaa Fadill Abbas
The Department of Information Networks at the College of Information Technology hosted the defense of a doctoral dissertation entitled “Network Slicing and Efficient Resource Management in SDN for Traffic Classification Using an Optimized Deep Neural Network Algorithm” by PhD candidate Rasool Azeem Mousa Abd, under the supervision of Prof. Dr. Ghaidaa Abdul-Hussein Bilal. The defense took place at 9:00 a.m. on Thursday, September 11, 2025, in the department’s conference hall.
The dissertation addressed the critical challenge of resource allocation in Software-Defined Networks (SDN), particularly in the context of fifth-generation (5G) environments. Due to the dynamic, heterogeneous, and highly demanding nature of 5G traffic, efficient resource management requires solutions that optimize bandwidth utilization, minimize latency, and ensure scalability for diverse services and applications. To meet these challenges, the study proposed an integrated system for intelligent resource allocation and traffic classification in 5G network slicing, combining two innovative approaches: Segmented Network Traffic Management (SNTM) and Weighted Network Traffic (WNT).
The SNTM approach enhances resource distribution across virtual network slices that share a common physical infrastructure, optimizing bandwidth utilization. The system was implemented using the RYU controller and Mininet emulator, where dynamic allocation significantly outperformed static allocation. Experimental results demonstrated that dynamic allocation achieved an average data throughput of 52.6 MB (compared to 48.23 MB), an average bandwidth of 60.87 Mbps (compared to 55.93 Mbps), and a packet loss reduction of 29.34% (from 21.1% to 14.913%), while maintaining acceptable jitter levels (0.301 ms versus 0.225 ms).
The second approach, WNT, incorporated an optimized Deep Neural Network (DNN) to achieve highly accurate traffic classification. By applying advanced preprocessing techniques such as feature engineering and class reduction, network flows were classified into high, medium, and low weights based on bandwidth consumption. Evaluations conducted using the Labeled Flow Data Set for 75 Applications confirmed remarkable performance, with classification accuracy reaching 99.89%, a loss function value of 0.0043, and a training duration of 1 hour and 11 minutes across different data splits (70–30, 80–20, and 60–40).
Together, the integrated SNTM–WNT framework provides a robust solution for resource allocation challenges in SDN. By intelligently managing traffic, improving bandwidth efficiency, and enabling adaptive slicing, the proposed framework contributes significantly to advancing network slicing methodologies in 5G environments through scalable and dynamically adaptable control strategies.