Traffic Prediction-Based Routing in Software-Defined Networking Using Long Short-Term Memory and Harmony Search

Efficient traffic management in Software-Defined Networking (SDN) relies heavily on accurate traffic prediction to enable intelligent routing decisions. In this paper, we propose a novel routing framework that integrates deep learning-based traffic forecasting with an optimization-driven path selection mechanism. Specifically, Long Short-Term Memory (LSTM) networks are employed to predict future traffic patterns based on time-series data extracted from SDN switches. To enhance prediction accuracy, Fast Fourier Transform (FFT) is used to enrich the input features with dominant traffic harmonics. Furthermore, key forecasting parameters such as the number of lags, labels, and shifts are optimized using the Harmony Search (HS) algorithm. The predicted traffic volumes are then utilized to dynamically adjust routing paths, aiming to minimize latency, jitter, and packet loss while improving throughput. The proposed framework is implemented on a Mininet-based SDN testbed using a customized Ryu controller. Experimental results under both regular and chaotic traffic scenarios demonstrate that the LSTM-based forecasting significantly improves routing efficiency compared to traditional approaches, especially when enhanced by FFT preprocessing and HS optimization. This study highlights the potential of combining predictive intelligence and metaheuristic tuning to achieve adaptive and robust traffic-aware routing in SDN environments.