Anomaly Detection in Mobile Broadband Network Using Federated Learning

Abstract

The advancement of mobile broadband networks has introduced new challenges in data privacy and the efficiency of anomaly detection. Traditional centralized data processing methods are becoming less effective, calling for innovative approaches like federated learning. This master’s thesis investigates the application of federated learning to enhance anomaly detection within mobile networks by utilizing real-world data from multiple locations.

The objective is to evaluate its ability to train locally on devices, thereby preserving user privacy and leveraging the data’s geographical diversity.

The research methodology involved training machine learning models directly on network nodes without centralizing the data. Over 20 rounds of training, the models demonstrated substantial improvements in detecting network anomalies. Notably, the best-performing round achieved an accuracy of 86.91%, a precision of 81.18%, and a consistently high recall of over 99%, resulting in an F1 score of 89.17%. These results confirm the effectiveness of federated learning in maintaining high detection rates while enhancing data privacy.

The findings highlight federated learning’s potential to transform network management practices, facilitating secure anomaly detection across different geographical regions.

The thesis concludes with a discussion on the scalability of federated learning models and their potential integration with existing technologies to further improve network functionality and security.