Multi-Agent Intelligent Architectures Enabling Scalable, Predictive, And Self-Healing IoT Ecosystems
Year: 2026
Type: conference-paper
Abstract: The fast development of Internet of Things (IoT) ecosystems requires smart and autonomous methods that ensure scalability, fault-tolerance, and uninterrupted functioning. The current research is aimed at the creation of a Multi-Agent Intelligent Architecture for predictive maintenance, anomaly detection, and self-healing in IoT ecosystems. The offered model includes a set of decentralized software agents like data collecting, predictive maintenance, anomaly detection, and self-healing agents that allow to make decisions and recover autonomously. Hybrid IoT dataset with 50,000 sensor records was used, which included 38,500 normal data samples and 11,500 fault data samples in an 80:20 train-test ratio. Three machine learning algorithms, namely Random Forest, Support Vector Machine (SVM), and KNN, were applied and tested against the proposed model. The model demonstrated 91.5% accuracy, 88.3% precision, 90.7% recall, and 89.5% F1-score, which outperformed Random Forest with 85.7% accuracy and SVM with 83.2% accuracy. The analysis of contributions of each agent proved the significance of each agent by reducing accuracy up to 86.9%, 88.2%, and 89.0% when eliminating predictive maintenance, anomaly detection, and self-healing agents, respectively. This architecture will facilitate improved autonomous fault management capabilities, increase IoT reliability, and offer a viable method for constructing reliable IoT systems. Future research will be centered around implementing advanced techniques of reinforcement learning.
Authors: PriyaViji, Utkarsh Anand
Institution: Kalinga University
Language: English
Topic: Internet of Things and AI
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7224924902