Machine Learning-Based in-Car KPI Predictions for Cellular Networks

The swift rollout of 4G and 5G networks presents challenges in monitoring and optimizing network performance, especially in dynamic vehicular environments. Key Performance Indicators (KPIs) for coverage and quality play crucial role in assessing network performance. However, due to mobility constraints and measurement limitations, KPI data collection is often difficult or incomplete. This paper explores machine learning-based approaches to predict missing KPI values and estimate network performance at unmeasured positions inside a vehicle. Measured data for three operators with test devices positioned at different locations within a moving vehicle were preprocessed before training four machine learning models: CatBoost, Extra Trees, LightGBM, and XGBoost. Model evaluation showed Extra Trees achieving the highest accuracy for the KPI estimation, while LightGBM demonstrated the most consistent performance across vehicle positions. Results indicated that front vehicle positions exhibit stronger but more variable signals, while back positions provide more stable yet weaker signals. The analysis highlights the importance of position-aware AI models for network optimization in vehicular settings.