Keywords
ML, Cellular Nets, Handover, Prediction, Rail

Technologies
Python, Pandas, Scikit-learn, PyTorch, LTE

Reliable mobile connectivity on moving trains remains a challenge due to frequent cellular handovers caused by rapidly changing network conditions. This project explores if machine learning can be used to predict future handover events, with the aim of improving on-board internet connectivity for passengers by facilitating more intelligent network management.

A dataset was created using real world measurements collected on the Dublin Area Rapid Transport (DART). Cellular network signal metrics were combined with spatial and temporal features derived from GPS data. A Mixture-of-Experts (MoE) model was chosen for its suitability to the heterogeneous nature of rail routes, capturing varying network behaviour across different locations and conditions. This allowed the system to specialise in distinct connectivity patterns observed along this route.

The current results indicate that handover events and link quality exhibit strong spatial structure, and that predictive models could leverage this to anticipate future network conditions. This work provides a foundation for more reliable and intelligent connectivity in high-mobility environments.