Efficient Network Traffic Prediction: Harnessing Hybrid Neural Networks

The prediction of network traffic plays a crucial role in Software-Defined Networking (SDN) for proactive management and dynamic traffic handling. However, obtaining accurate historical data remains a challenge in traditional networks. This study proposes a hybrid machine learning model that combines Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory networks (Bi-LSTM) for traffic load prediction. The model leverages both historical and real-time network traffic data to forecast future traffic dynamics. Evaluation on GEANT and Albein datasets demonstrates that the proposed model outperforms established methodologies including GRU, RNN, LSTM, and BiLSTM, with improvements in Root Mean Squared Error (RMSE) ranging from 35.2% to 503% and Mean Absolute Error (MAE) ranging from 22.2% to 64%. These findings underscore the efficacy of the hybrid model in enhancing predictive accuracy within SDN frameworks, thereby offering insights for more efficient network management practices.