Original Contribution to Differential Privacy Mechanisms for Federated Learning in Telecommunications Network Intelligence Applications in Nigeria

Federated learning enables distributed machine learning model training across multiple network operator data sources without centralized data collection, making it technically suitable for collaborative network intelligence applications where data sovereignty constraints prevent data pooling. However, federated learning without privacy guarantees remains vulnerable to inference attacks that can reconstruct sensitive subscriber data from shared model updates. This doctoral study develops original differential privacy mechanisms for federated learning in telecommunications network intelligence applications, addressing the specific privacy and utility trade-offs relevant to Nigerian multi-operator collaborative learning scenarios. An original privacy-preserving federated learning framework incorporating adaptive Gaussian noise calibration based on per-round gradient sensitivity analysis is developed, providing formal epsilon-differential privacy guarantees while minimizing the utility degradation associated with fixed-noise mechanisms. The framework is validated in two Nigerian telecommunications application scenarios: collaborative network anomaly detection across three simulated operator networks, and joint quality of experience prediction for OTT video across two ISP networks sharing anonymized session data. Privacy analysis demonstrates that the proposed mechanism achieves epsilon values of 0.8 for the anomaly detection application, meeting the target privacy budget while preserving 94.1% of the model accuracy achievable without privacy constraints. The federated anomaly detection model identifies network security incidents with 91.3% detection rate, outperforming individual operator models by 17 percentage points due to the broader attack pattern exposure from multi-operator training. The original differential privacy mechanism constitutes a contribution to both theoretical privacy literature and applied telecommunications intelligence. Keywords: federated learning, differential privacy, network intelligence, anomaly detection, Nigeria.