Multimodal AI Boosts Cellular Traffic Forecasting Accuracy.

Accurate forecasting of cellular network traffic is vital for efficient network planning, resource allocation, and maintaining quality of service in modern mobile communication systems. However, real-world traffic is complex, exhibiting bursty dynamics and being influenced by external urban events, which makes reliable prediction challenging. Most existing spatiotemporal traffic forecasting methods tend to focus on intrinsic traffic patterns or structural relationships within a single data modality, often neglecting burst behavior and exogenous contextual signals. To address these limitations, researchers propose MSPF-Net, a multimodal cellular traffic forecasting framework. MSPF-Net integrates various information sources through several modules: a Spatiotemporal-Frequency Traffic Encoder captures temporal, spatial, and spectral traffic patterns; a Peak Enhancement Module extracts burst-aware representations of sudden spikes; a News Context Representation Module encodes urban news streams into contextual embeddings; and a Dynamic Fusion Prediction Module adaptively combines these diverse signals for accurate forecasts. Experiments on real-world datasets like Milano, Trento, and LTE traffic demonstrate that this joint modeling approach significantly enhances forecasting performance.