5G Radio Cell Performance Prediction Based on Knowledge Graph Fusion with Neural Network

Existing data-driven methods for predicting 5G cell performance do not use expert knowledge effectively. This paper presents a methodology for the integration of domain expert knowledge on cell configuration dependencies, represented as a knowledge graph, into a regression model. Evaluation with live 5G network data and a state-of-the-art predictor indicates that the proposed approach outperforms the reference model and allows identification of configuration elements with the most significant impact on performance.