Geographic ML Model Selection for Subscriber Experience Prediction

The patent divides the serving cell into multiple geographic areas or zones, each with its own machine learning model. This segmentation allows the system to capture local environmental characteristics and external factors specific to each area, thereby improving prediction accuracy without requiring a single overly complex model to account for all variations across the entire cell.

The patent implements location-specific machine learning models that are trained on data from particular geographic areas within the serving cell. Each model learns the local patterns, environmental conditions, and external factors relevant to its specific area, enabling more accurate predictions of subscriber experience for UEs in that location rather than using a generalized cell-wide model.

The patent implements a dynamic model selection mechanism that adapts to changing conditions. The system monitors various parameters such as UE location, environmental conditions, and model performance metrics, and dynamically selects or switches between different machine learning models based on current conditions. This dynamic approach allows the system to maintain high prediction accuracy while managing complexity by only activating or selecting models when and where they are needed.

The system includes automated model selection and performance evaluation mechanisms that operate without manual intervention. The machine learning framework automatically evaluates multiple models, selects the most appropriate one for given conditions, and can even retrain or update models based on new data. This self-service capability reduces the operational complexity of managing multiple location-specific models while maintaining their predictive benefits.