Wrocław University of Science and Technology is responsible for the methodological layer of the project, including the development and validation of statistical and machine-learning methods that underpin the automated assessment of how configuration changes affect network performance.

The research carried out by the Faculty of Mathematics is classified as industrial research and is based on anonymized measurement data provided by the Consortium Leader, including network performance indicators and data describing the network's operating conditions.

The work focuses on three closely related areas.

The first is the automation of domain-specific predictive models for performance indicators, including the selection of explanatory variables, model training and validation procedures, and the assessment of predictive uncertainty. The aim is to make it possible to develop a model for a new domain in a repeatable way, without requiring expert input each time.

The second involves methods for improving the quality of these models by making use of data at different resolutions and levels of aggregation, including model fine-tuning, ensemble methods, and prediction fusion.

The third involves methods for representative sampling of multidimensional performance and environmental characteristics. These methods make it possible to identify a limited set of network operating conditions that reliably reflects the network's actual variability.

The work will result in documented procedures, computational prototypes, and methodological reports, validated against real-world data according to the accuracy criteria specified in the funding application.

The results will be delivered to the Consortium Leader and used in the development of the network's digital twin and a recommendation system for configuring base-station functionalities. The collaboration will also create opportunities for scientific publications and for developing undergraduate and graduate thesis topics based on data from a real telecommunications network.