Intelligent 5G Network Performance Optimization through Gradient Boosting

Performance assurance in fifth-generation (5G) networks is increasingly challenging due to heterogeneous service requirements and rapidly varying radio and traffic conditions. This paper presents an interpretable supervised-learning framework for 5G network performance classification under high-traffic behaviour. Using a 5G KPI dataset with 1000 samples and 25 indicators [dimartino2020_5g_metrics], we introduce domain-informed composite metrics---including the Signal Quality Index (SQI), Network Efficiency Ratio (NER), QoS Performance Index (QPI), and Energy Traffic Ratio (ETR)---and augment the input with additional engineered features (33 features in total). We benchmark ten classifiers under three train--test splits (80:20, 60:40, and 50:50) using accuracy, precision, recall, F1-score, and AUC--ROC. Gradient Boosting provides the best overall accuracy (99.0%) and AUC--ROC (99.6%). Feature-importance analysis shows that latency and QPI dominate the model decisions (approximately 69% cumulative importance), highlighting actionable QoS levers during high-load operation.