Forecasting Network Interface Flow Using a Broad Learning System Based on the Sparrow Search Algorithm.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Network Engineering

Background:

• Broad learning systems (BLS) offer efficient learning but require careful hyperparameter tuning.
• Network traffic forecasting is crucial for resource management and performance optimization.
• Manual hyperparameter optimization can be time-consuming and suboptimal.

Purpose of the Study:

• To enhance the prediction accuracy of broad learning systems for network traffic forecasting.
• To automate and optimize the selection of hyperparameters (shrinkage and regularization coefficients) in BLS.
• To develop a robust network interface flow forecasting model using an optimized BLS.

Main Methods:

• Utilizing the sparrow search algorithm (SSA) to optimize the shrinkage (r) and regularization (λ) coefficients of the broad learning system.
• Employing historical network flow data (time period [T-11,T]) as features for predicting traffic at moment T+1.
• Training and validating the proposed model on two public network flow datasets and a real-world enterprise cloud platform dataset.

Main Results:

• The SSA-optimized BLS demonstrated superior prediction accuracy compared to standard BLS, long short-term memory (LSTM), and other baseline methods.
• The model achieved high moving average accuracy, reaching 97%, 98%, and 99% on the tested datasets.
• Automated hyperparameter optimization via SSA effectively improved the BLS performance.

Conclusions:

• The proposed sparrow search algorithm-based broad learning system offers a highly accurate and efficient approach for network traffic forecasting.
• Automated hyperparameter optimization is a key factor in maximizing the performance of broad learning systems.
• The model's effectiveness is validated across multiple datasets, indicating its practical applicability in real-world network environments.