FEDERATED LEARNING-BASED DATA COLLABORATION METHOD FOR ENHANCING EDGE CLOUD AI SYSTEM SECURITY USING LARGE LANGUAGE MODELS

Abstract

The rapid growth of Edge Computing, Internet of Things (IoT), Cloud Computing, and Artificial Intelligence (AI) has significantly increased the volume of distributed data generated across heterogeneous devices. Traditional centralized machine learning approaches require transferring raw data to cloud servers for model training, creating challenges related to data privacy, security, communication overhead, and regulatory compliance. Federated Learning (FL) has emerged as an effective privacy-preserving paradigm that enables collaborative model training without exposing sensitive local datasets. However, existing federated learning frameworks remain vulnerable to model poisoning attacks, malicious participant behavior, and insufficient trust management, which can degrade global model performance and compromise system reliability. To address these challenges, this paper proposes a Federated Learning-Based Data Collaboration Method for Enhancing Edge Cloud AI System Security Using Large Language Models (LLMs). The proposed framework integrates Federated Averaging (FedAvg), Isolation Forest-based anomaly detection, dynamic trust management, and Large Language Models to establish a secure, scalable, and intelligent edge-cloud AI ecosystem. Edge devices independently train local machine learning models using private datasets, while only encrypted model parameters are shared with the central aggregation server. The Isolation Forest algorithm continuously monitors incoming model updates to identify malicious or anomalous behavior, while a trust evaluation mechanism dynamically assigns trust scores to participating nodes based on historical contributions and detected anomalies. Furthermore, an LLM-based intelligence module analyzes security events, interprets detected threats, and generates human-readable cybersecurity reports with actionable mitigation recommendations, thereby improving transparency and decision-making. The framework is implemented using Python, Django, MySQL, Scikit-Learn, Hugging Face Transformers, HTML, CSS, JavaScript, and Chart.js. Experimental evaluation demonstrates improved privacy preservation, enhanced anomaly detection accuracy, reduced communication overhead, robust trust management, and explainable cybersecurity analytics compared with conventional federated learning architectures. The proposed system provides a reliable and intelligent solution for securing distributed artificial intelligence infrastructures in healthcare, industrial IoT, smart cities, autonomous systems, and next-generation edge-cloud computing environments.