A data-Driven Approach To Using Artificial Intelligence To Improve Network Communication Efficiency
Year: 2026
Type: article
Abstract: The rapid growth of data-driven communication networks has increased the need for accurate network traffic analysis and throughput prediction to support efficient resource utilization and intelligent network management. However, conventional prediction methods often struggle to capture complex temporal dependencies, nonlinear variations, and dynamic traffic patterns, resulting in limited forecasting accuracy and generalization. This study proposes a deep learning-based network throughput prediction framework using the CESNET-TimeSeries24 dataset, which contains large-scale, regularly sampled network traffic statistics. The framework incorporates data preprocessing, including missing-value handling, outlier treatment, feature normalization, and sliding-window transformation for effective temporal modeling. A Bidirectional Long Short-Term Memory (BiLSTM) network is employed to learn both short- and long-term traffic dependencies from forward and backward temporal sequences. Experimental results demonstrate strong predictive performance, achieving an mean absolute error (MAE) of 0.0306, mean squared error (MSE) of 0.0018, and root mean squared error (RMSE) of 0.0429. The findings confirm that the proposed framework provides accurate and consistent throughput predictions. These predictions may potentially support intelligent traffic analysis, efficient resource allocation, and network performance optimization in modern high-speed communication networks; however, these downstream network management benefits are not directly evaluated in the present study.
Author: YuSheng Jiang
Institution: Shaanxi University of Science and Technology
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7237344355
FWCI:
Cites:
Cited by:
Related to:
