Answers are grounded in this paper and cite their sources.
Software-Defined Networking for the Internet of Things (SDN-IoT), as an emerging network architecture, faces urgent challenges in dynamic resource allocation and network congestion risks. Fortunately, refined network traffic prediction can model traffic evolution patterns to enable dynamic resource allocation and congestion early warning. Nevertheless, existing traffic prediction methods primarily target traditional Internet Service Content (ISC) networks and face three critical limitations in SDN-IoT scenarios: First, insufficient differentiation between intra- and inter-period temporal feature variations leads to inadequate modeling of cross-period evolution patterns, significantly degrading prediction accuracy. Second, centralized training architectures struggle to adapt to distributed network environments , with data silos and privacy protection concerns. Third, noise features in scenarios like network congestion lack effective filtering, compromising feature extraction reliability. To address these challenges, focuses on the core perspective of ”intra- and inter-period feature fusion” and proposes DRAFT - a high-precision federated learning method for SDN-IoT network traffic prediction. The approach first designs a Temporal Optimization TimesNet that identifies significant periods through Fast Fourier Transform (FFT), constructing Conv1 convolution modules to capture intra- and inter-period variation features, achieving collaborative dual-domain feature representation and prediction performance enhancement. Subsequently, a federated learning framework is established to implement traffic path planning and access control strategies, resolving data silos and privacy issues. Finally, a robust denoising mechanism employing Empirical Mode Decomposition (EMD) and adaptive Wavelet Threshold (WT) filtering is developed to enhance feature extraction quality. Experiments conducted on two real-world network traffic datasets-UK_Academic_Dataset and EU_Core _Dataset demonstrate the superior predictive capabilities of DRAFT. • Design Temporal Optimization TimesNet to capture multi-domain features. • Built a federated learning framework DRAFT on SDN-IoT for network traffic prediction. • Design robust denoising mechanisms in DRAFT to eliminate noise in traffic data. • DRAFT’s performance improves by up to 95.28 percent over other advanced work. • DRAFT enables traffic prediction with SDN path planning to enhance network defence.
No takes yet. Share an insight, caveat, or question.
Yan et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: