Neural network approaches for quality-of-service optimization in software-defined networking environments

Abstract

Software-defined networking (SDN) enables centralized and programmable control of network behavior; however, conventional routing strategies remain largely reactive and struggle to adapt to rapidly changing traffic dynamics. To address this limitation, this study proposes a learning-based SDN routing framework that integrates a long short-term memory (LSTM) model to predict traffic patterns and proactively optimize routing decisions. The proposed approach is implemented and evaluated in an SDN testbed using realistic traffic scenarios. Experimental results are averaged over multiple independent runs to ensure robustness and reproducibility. Compared with static shortest-path routing and classical machine learning (ML) baselines, the proposed model demonstrates consistent improvements in latency, packet loss, and throughput under the evaluated conditions. In particular, the ablation study reports a 95% confidence interval for end-to-end latency ranging from 51.8 to 55.6 ms, confirming the statistical stability of the observed gains. Additional analyses show that the framework maintains low inference latency and modest control overhead, making it suitable for real-time SDN environments. Overall, the findings indicate that temporal learning models can effectively enhance SDN routing performance when evaluated within controlled experimental settings, offering a practical pathway toward more adaptive and intelligent network control.