Communication Dans Un Congrès Année : 2025

Large Language Models for Spatio-Temporal Mobile Traffic Predictions

Résumé

Reliable mobile network traffic prediction is an a priori requirement for most of the network planning and optimization tasks. In this paper, we exploit Large Language Models (LLM) for traffic prediction, and explore their ability for integrating the spatio-temporal correlation. Our results indicate that when LLMs are fine-tuned using traffic data that includes both spatial and temporal features, they significantly outperform traditional predictive models in terms of accuracy. Furthermore, we highlight that this fine-tuning process necessitates only a limited amount of data, thereby rendering it a viable approach for various applications in the field.

Fichier principal
Vignette du fichier
m35855-madi final.pdf (1.24 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05109672 , version 1 (12-06-2025)

Licence

Identifiants

  • HAL Id : hal-05109672 , version 1

Citer

Ryad Madi, Salah-Eddine Elayoubi, Ricardo Chiquetto Do Lago, Vitor Opsfelder Estanislau. Large Language Models for Spatio-Temporal Mobile Traffic Predictions. Workshop on Next-Gen Networks through LLMs, Action Models, and Multi-Agent Systems (NextGenAI) at the fifth IEEE International Mediterranean Conference on Communications and Networking (IEEE MEDITCOM 2025), Jul 2025, Nice, France. ⟨hal-05109672⟩
176 Consultations
633 Téléchargements

Partager

  • More