Multi-Step Internet Traffic Forecasting Models with Variable Forecast Horizons for Proactive Network Management

Area of Science:

• Computer Science
• Network Engineering
• Data Science

Background:

• Internet Service Provider (ISP) networks require accurate internet traffic forecasting (ITF) for strategic planning and network management.
• Existing ITF methods struggle with real-world aberrant data, leading to network congestion and over-provisioning.
• Proactive network management necessitates robust forecasting models capable of handling data anomalies.

Purpose of the Study:

• To develop and evaluate an innovative ITF model for proactive network management in ISP environments.
• To address the limitations of traditional ITF models by incorporating outlier detection and mitigation.
• To enhance the accuracy and robustness of internet traffic predictions using real ISP data.

Main Methods:

• The proposed model integrates outlier detection and mitigation with gradient descent and boosting algorithms (GBR, XGB, LGB, CBR, SGD).
• The model was evaluated using real internet traffic data from high-speed ISP networks.
• Performance was assessed across multiple forecast horizons (6, 9, and 12 steps).

Main Results:

• The developed ITF model demonstrated superior predictive accuracy compared to traditional forecasting models.
• The integration of outlier detection and mitigation significantly improved model performance and robustness.
• The model proved adaptable and effective across different forecast horizons.

Conclusions:

• The novel ITF model offers a significant advancement for proactive network management in ISP industries.
• Accurate forecasting, especially with outlier handling, is crucial for preventing network congestion and optimizing resource allocation.
• The model's effectiveness on real-world ISP data validates its practical applicability.