Log Anomaly detection in IoT networks using DeBERTa-V3
Haralambie Florin, Bratu Alexandru, Purcăruș Raluca-Ioana, Goga Nicolae, Andrei Vasilăţeanu · 2025
In the ever-changing environment of IT security, the need for having as much data about each appliance in the network, at any time, is dire. For this reason, in this paper, a monitorization system is described, that feeds on common data for almost all device types, more precise logs. Logs are an inherent feature of computer systems, but are an especially useful resource for IoT systems because of their rapid adoption in various domains. Preceded by a short literature review, a discussion over the use cases of such a system is held, followed by a description of the logs structure and learning scenarios. A Python script for simulating custom logs was developed, that generates data for training the AI model, both supervised and semi-supervised, based on DeBERTa-V3, a promising architecture with great potential for supervised and semisupervised learning.