Proactive Network Traffic Prediction using Generative Adversarial Network

Gyurin Byun, Van Vi Vo, Syed M. Raza, Duc Tai Le, Huigyu Yang, Hyunseung Choo

Abstract: Proactive traffic management in 5G networks is crucial for optimizing network efficiency, ensuring quality of service, adapting to dynamic conditions, and enhancing the user experience. In this paper, we propose a Generative Adversarial Network (GAN) architecture that leverages spatiotemporal features in network traffic data to predict future traffic. Our approach incorporates a Convolutional Long Short-Term Memory (ConvLSTM) model within the generator of the GAN, which collaborates with the discriminator, utilizing a Convolutional Neural Network (CNN) model, to provide essential feedback for training the generator. This integration ensures that our model not only predicts future traffic with improved accuracy but also adapts to dynamic network conditions. Based on experimental results using network traces, our model significantly outperforms the baseline, reducing prediction error by 12% while forecasting network traffic for the next 1 minute. These findings represent a significant advancement in proactive network management, particularly in addressing the challenges posed by real-time streaming and other latency-sensitive applications in 5G networks.