Building upon my previous work, my current objective is to create a general and adaptable approach to anomaly detection in Software-Defined Internet of Things (SD-IoT) networks, leveraging the capabilities of In-Band Network Telemetry (INT). Unlike conventional methods tailored for specific scenarios, our approach ensures broad applicability across diverse SD-IoT configurations while dynamically adapting to evolving network conditions.

Another key novelty is using INT data instead of traditional network datasets collected from tools like Wireshark and NetFlow. INT, a real-time telemetry mechanism, provides granular, direct data plane insights—such as hop latency, flow latency, and queue depth—offering a richer and more dynamic foundation for anomaly detection. To achieve robust detection, I developed neural network models that integrate theoretical insights with recent advances in machine learning. Additionally, we incorporated advanced techniques, such as ensemble learning and attention mechanisms, to enhance generalizability and adaptability.