Llm-Augmented Cyber Threat Detection Using Federated Edge Intelligence and Data Mining Techniques
Mohammed Shakeer Bandrevu, NagaSatyanarayana Raju Uppalapati, Ramesh Bellamkonda, Srikanth Kamatala · 2025
Due to the rising complexity of cyber threats across digital networks, we require smart and decentralized methods that respect privacy for catching those threats. A new method for finding cyber threats is proposed here, using Large Language Models in combination with Federated Edge Intelligence and modern Data Mining. LLMs are used by the proposed system to add depth to the semantic analysis of traffic and logs in the network. By using federated learning, model training can occur on different gadgets together, with information kept safe and the need for data exchange reduced. The use of data mining enables the system to find valuable information and unusual cases in data sources at any moment. Thanks to LLM, ML models and pattern detection, the service is able to detect threats before they happen and still ensure user information is protected. The system proved better results in terms of accuracy, scalability and adaptability than the centralized and simpler models tested with NSL-KDD and CICIDS2017 datasets. The outcomes suggest that using LLM in federated frameworks makes edge environments safer from new cyber risk.