Comparative Analysis of Deep Learning Models for Real-World ISP Network Traffic Forecasting

2025-03-20

AI Summary

This work addresses the problem of accurate traffic forecasting in ISP networks to support resource optimization, user experience enhancement, and anomaly detection. Methodologically, we introduce the first reproducible, unified evaluation framework built upon the real-world multivariate time-series dataset CESNET-TimeSeries24—comprising 40 weeks of backbone and access-layer traffic. We systematically benchmark mainstream deep temporal models (LSTM, GRU, TCN, Transformer, Informer) across network granularity levels, employing standardized preprocessing, sliding-window training, and multi-step forecasting protocols. Our contributions include: (i) establishing a rigorous performance benchmark and comparative methodology for ISP traffic forecasting; (ii) quantifying accuracy-efficiency trade-offs; and (iii) revealing distinct model applicability boundaries between backbone and access layers—demonstrating up to a 12.7% reduction in MAE for the best-performing model. All code and configurations are publicly released.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Machine Learning: Time-Series/Data Streams
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterization
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Web Mining and Content Analysis: Web traffic and log analysis

Abstract

Accurate network traffic forecasting is essential for Internet Service Providers (ISP) to optimize resources, enhance user experience, and mitigate anomalies. This study evaluates state-of-the-art deep learning models on CESNET-TimeSeries24, a recently published, comprehensive real-world network traffic dataset from the ISP network CESNET3 spanning multivariate time series over 40 weeks. Our findings highlight the balance between prediction accuracy and computational efficiency across different levels of network granularity. Additionally, this work establishes a reproducible methodology that facilitates direct comparison of existing approaches, explores their strengths and weaknesses, and provides a benchmark for future studies using this dataset.

Problem

• Evaluates deep learning models for ISP traffic forecasting accuracy
• Compares computational efficiency across network granularity levels
• Establishes reproducible benchmark for future traffic prediction studies

Innovation

• Evaluates deep learning models on CESNET-TimeSeries24 dataset
• Balances prediction accuracy and computational efficiency
• Establishes reproducible methodology for model comparison