Generalizable IoT Traffic Representations for Cross-Network Device Identification

Machine learning models have demonstrated strong performance in classifying network traffic and identifying Internet-of-Things (IoT) devices, enabling operators to discover and manage IoT assets at scale. However, many existing approaches rely on end-to-end supervised pipelines or task-specific fine-tuning, resulting in traffic representations that are tightly coupled to labeled datasets and deployment environments, which can limit generalizability. In this paper, we study the problem of learning generalizable traffic representations for IoT device identification. We design compact encoder architectures that learn per-flow embeddings from unlabeled IoT traffic and evaluate them using a frozen-encoder protocol with a simple supervised classifier. Our specific contributions are threefold. (1) We develop unsupervised encoder--decoder models that learn compact traffic representations from unlabeled IoT network flows and assess their quality through reconstruction-based analysis. (2) We show that these learned representations can be used effectively for IoT device-type classification using simple, lightweight classifiers trained on frozen embeddings. (3) We provide a systematic benchmarking study against the state-of-the-art pretrained traffic encoders, showing that larger models do not necessarily yield more robust representations for IoT traffic. Using more than 18 million real IoT traffic flows collected across multiple years and deployment environments, we learn traffic representations from unlabeled data and evaluate device-type classification on disjoint labeled subsets, achieving macro F1-scores exceeding 0.9 for device-type classification and demonstrating robustness under cross-environment deployment.

IoT Behavioral Monitoring via Network Traffic Analysis

Smart homes, enterprises, and cities are increasingly being equipped with a plethora of Internet of Things (IoT), ranging from smart-lights to security cameras. While IoT networks have the potential to benefit our lives, they create privacy and security challenges not seen with traditional IT networks. Due to the lack of visibility, operators of such smart environments are not often aware of their IoT assets, let alone whether each IoT device is functioning properly safe from cyber-attacks. This thesis is the culmination of our efforts to develop techniques to profile the network behavioral pattern of IoTs, automate IoT classification, deduce their operating context, and detect anomalous behavior indicative of cyber-attacks. We begin this thesis by surveying IoT ecosystem, while reviewing current approaches to vulnerability assessments, intrusion detection, and behavioral monitoring. For our first contribution, we collect traffic traces and characterize the network behavior of IoT devices via attributes from traffic patterns. We develop a robust machine learning-based inference engine trained with these attributes and demonstrate real-time classification of 28 IoT devices with over 99% accuracy. Our second contribution enhances the classification by reducing the cost of attribute extraction while also identifying IoT device states. Prototype implementation and evaluation demonstrate the ability of our supervised machine learning method to detect behavioral changes for five IoT devices. Our third and final contribution develops a modularized unsupervised inference engine that dynamically accommodates the addition of new IoT devices and/or updates to existing ones, without requiring system-wide retraining of the model. We demonstrate via experiments that our model can automatically detect attacks and firmware changes in ten IoT devices with over 94% accuracy.