Original title: TD-STGT: A Spatio-Temporal Graph Transformer for Mobile Traffic Demand Forecasting
Mohamad Alkadamani, Halim Yanikomeroglu · September 2026
Planning future 5G and 6G cellular networks requires knowing where mobile traffic will grow. Phone companies need this to build new sites and upgrade equipment. To solve this, researchers built TD-STGT, a computer forecasting model. It predicts wireless mobile traffic demand changes across fine geographic grids. The model works by tracking a demand proxy that scales crowdsourced phone measurements with daytime population data. It then analyzes how these patterns change across both space and time. Tested across five Canadian metropolitan regions, TD-STGT outperformed existing models. It achieved a $ΔR^2$ score of 0.462 and lowered $Δ$RMSE by 5.7% compared to the strongest baseline. This tool helps network planners spot high-demand areas and prioritize where to add cellular capacity.
Imagine you are deciding where to set up food stalls at a large festival. You want to be where the crowd is, but the crowd moves unpredictably. Checking one spot only tells you about that moment, not where people will be later. In this paper the model predicts mobile data demand across city grids by learning how demand changes over time and space, similar to predicting crowd movement from past patterns and current density.
Network operators need to know where to upgrade next. This model predicts which small city blocks will see rising mobile data demand, so they can invest in radio upgrades and backhaul expansion exactly where needed. It was tested across five Canadian cities and outperformed other forecasting methods, making it a practical tool for planning future networks.
The work does not establish performance outside of five Canadian metropolitan regions. It does not address real-time forecasting or individual user behavior. The reliance on a population-scaled demand proxy may not capture all demand drivers, and the model's generalizability to other network types or data sources is not demonstrated.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
TD-STGT: A Spatio-Temporal Graph Transformer for Mobile Traffic Demand Forecasting — Mohamad Alkadamani, Halim Yanikomeroglu
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.