Answers are grounded in this paper and cite their sources.
In mobile networks, intrusion detection system needs to be able to detect anomalous traffic while not revealing any information that is not related to the subscribers. In this paper, we investigate a prototype of a project called AEGIS, which is an integration of a deep neural network (DNN) classifier and a browser-based 4G/5G monitoring interface, and create a privacy-preserving distributed extension of this prototype. Twenty flow features are included in the provided software; it can predict normal or attack traffic, adds network context based on rule sets, and keeps an alert history that is limited. The datasets provided are confirmed by an audit to have a total of 2,002 balanced records, with 1,200 training, 401 validation and 401 test samples. The model-comparison artifacts present an accuracy of 97.26% and attack recall of 99.00%, while beside the application-loaded model artifacts show an accuracy of 94.76% and an attack recall of 93.00%. These records are not considered to be one single verified experiment. The suggested extension continues to keep data on flows in participating sites and includes simultaneously local training, secure aggregation and a separately provided differential-privacy mechanism. These distributed controls and privacy controls are design proposals, not features provided by the distributed implementation. The study is an evidence based account of the prototype, detailing privacy exposure from external reputation services, and outlining experiments required to assess the proposed architecture.
No takes yet. Share an insight, caveat, or question.
G et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: