A Study on a Meta-Learning Network Traffic Forecasting Model Based on Gaussian Mixture Models
Year: 2025
Type: conference-paper
Abstract: Accurate forecasting of cellular traffic in non-stationary environments, where patterns dynamically emerge and vanish, remains a formidable challenge. To tackle this, we propose a novel meta-learning framework, GMM-SCM-DCM. It employs a Gaussian Mixture Model (GMM) for probabilistic meta-feature representation. The core innovation is a Dynamic Component Management (DCM) mechanism that enables online structural evolution of the meta-learner by dynamically splitting, merging, or pruning Gaussian components based on a bimodal similarity metric, ensuring sustained alignment with shifting data distributions. A Single-Component Mechanism (SCM) is utilized for precise base-learner initialization. To ensure realistic validation, we reconstructed the Telecom Italia Milan dataset via unsupervised clustering and meta-feature engineering to identify four distinct functional zones, including crucial non-stationary areas. Comprehensive experiments demonstrate that our model significantly outperforms traditional methods and meta-learning baselines, achieving a 9.3% reduction in MAE. The model’s superior generalization capability and practical utility are further confirmed through ablation studies, robustness tests, and successful cross-dataset validation on the Shanghai Telecom dataset.
Institution: Space Engineering University
Language: English
Subfield: Building and Construction
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7127042587