Spectrum-Aided Traffic Decomposition and Deep Learning Method for Network Traffic Prediction in Internet of Things

Network traffic prediction plays a crucial role in optimizing resource allocation, mitigating congestion, and enhancing cybersecurity in large-scale Internet of Things systems. However, the multiscale temporal patterns, complexity, and nonlinearity of traffic data pose significant challenges for accurate modeling and prediction. To address these limitations, this article proposes a novel approach that combines spectrum-aided traffic decomposition with the deep learning (DL) method for network traffic prediction. Specifically, spectrum-aided traffic decomposition is performed using the seasonal-trend decomposition using loess algorithm to decompose traffic data into interpretable seasonal, trend, and residual components, where the seasonal period is dynamically optimized through frequency domain analysis. Each component is then modeled separately using a DL architecture, which is the gated recurrent units-based sequence-to-sequence model with an attention mechanism. This allows the model to effectively capture multiscale temporal dependencies, long-term relationships, and complex patterns, thereby enhancing prediction accuracy. The predictions from each decomposed component are subsequently ensembled to obtain the final prediction value. Experimental results show that our method achieves significant improvements over other methods across four real-world datasets, reducing mean square error and mean absolute error by an average of 97.6% and 84.7%, respectively.