DIFT-VAR: A DYNAMIC MULTI-LAYER FRAMEWORK FOR DEVICE FINGERPRINTING OF IDENTICAL DEVICES

 Device fingerprinting is a powerful technique for identifying devices in an IoT environment, offering multiple advantages such as enhanced security through device authentication, improved network management by monitoring device behaviors, and anomaly detection for identifying unauthorized or compromised devices. The majority of recent fingerprinting schemes consider a heterogeneous device environment and use different machine learning techniques to identify devices using network traffic, signal-level information, radio frequency characteristics, etc. However, fingerprinting devices of the same make and model is a significant challenge in modern IoT environments, where many devices often share identical hardware and software configurations. Existing techniques cannot reliably differentiate identical devices as they lack sufficient data. This paper proposes a novel approach for Device Identification and Fingerprinting with Time-Variant Adaptive Recognition (DIFT-VAR) based on multi-layer, time-varying feature extraction. We construct dynamic fingerprints that uniquely identify each device by monitoring and fusing features such as probe request behavior, clock skew, transport layer characteristics, and radio signal metrics over time. We utilize machine learning algorithms such as Random Forests to classify devices based on these dynamic fingerprints. We further propose the use of dynamic time warping (DTW) for feature alignment and classification. Experimental results demonstrate the efficacy of our approach in distinguishing identical devices with an accuracy of over 97% using standard machine learning metrics.