• español
    • English
  • Login
  • español 
    • español
    • English
  • Tipos de Publicaciones
    • bookbook partconference objectdoctoral thesisjournal articlemagazinemaster thesispatenttechnical documentationtechnical report
Ver ítem 
  •   IMDEA Networks Principal
  • Ver ítem
  •   IMDEA Networks Principal
  • Ver ítem
JavaScript is disabled for your browser. Some features of this site may not work without it.

A Scalable DNN Training Framework for Traffic Forecasting in Mobile Networks

Compartir
Ficheros
authors' version.pdf (2.685Mb)
Identificadores
URI: https://hdl.handle.net/20.500.12761/1908
Metadatos
Mostrar el registro completo del ítem
Autor(es)
Moghadas Gholian, Serly; Fiandrino, Claudio; Widmer, Joerg
Fecha
2025-05
Resumen
The exponential growth of mobile data traffic demands efficient and scalable forecasting methods to optimize network performance. Traditional approaches, like training individual models for each Base Station ( BS) are computationally prohibitive for large-scale production deployments. In this paper, we propose a scalable Deep Neural Networks (DNN) training framework for mobile network traffic forecasting that reduces input redundancy and computational overhead. We minimize the number of input probes (traffic monitors at Base Stations (BSs)) by grouping BS s with temporal similarity using K-means clustering with Dynamic Time Warping (DTW ) as the distance metric. Within each cluster, we train a DNN model, selecting a subset of BSs as inputs to predict future traffic demand for all BSs in that cluster. To further optimize input selection, we leverage the well-known EXplainable Artificial Intelligence ( XAI) technique, LayeR-wise backPropagation ( LRP) to identify the most influential BS s within each cluster. This makes it possible to reduce the number of required probes while maintaining high prediction accuracy. To validate our newly proposed framework, we conduct experiments on two real-world mobile traffic datasets. Specifically, our approach achieves competitive accuracy while reducing the total number of input probes by approximately 81% compared to state-of-the-art predictors.
Compartir
Ficheros
authors' version.pdf (2.685Mb)
Identificadores
URI: https://hdl.handle.net/20.500.12761/1908
Metadatos
Mostrar el registro completo del ítem

Listar

Todo IMDEA NetworksPor fecha de publicaciónAutoresTítulosPalabras claveTipos de contenido

Mi cuenta

Acceder

Estadísticas

Ver Estadísticas de uso

Difusión

emailContacto person Directorio wifi Eduroam rss_feed Noticias
Iniciativa IMDEA Sobre IMDEA Networks Organización Memorias anuales Transparencia
Síguenos en:
Comunidad de Madrid

UNIÓN EUROPEA

Fondo Social Europeo

UNIÓN EUROPEA

Fondo Europeo de Desarrollo Regional

UNIÓN EUROPEA

Fondos Estructurales y de Inversión Europeos

© 2021 IMDEA Networks. | Declaración de accesibilidad | Política de Privacidad | Aviso legal | Política de Cookies - Valoramos su privacidad: ¡este sitio no utiliza cookies!