PowerPrint: Harnessing Machine Learning for Accurate and Scalable Device Fingerprinting via Power Consumption

This paper examines the application of machine learning techniques for enhanced device fingerprinting based on power consumption. Existing works on device identification in large-scale networks struggle with accurately pinpointing individual machines among many identical devices, often relying on static fingerprinting or substantial human-involved decision-making, which is inefficient for real-time environments. In this work, we propose PowerPrint, a machine learning-based methodology designed to enhance device fingerprinting by analyzing power consumption data. PowerPrint leverages multiclass classification models to first detect two candidate machines, then further narrows down the identification to a single machine. This two-step online inference process, driven by power consumption analysis, enables precise identification within predefined categories, ensuring accurate and efficient device fingerprinting in large-scale networked systems. Experimental results indicate the effectiveness of our novel approach, with ExtraTrees and MLP classifiers achieving F1-scores of 95% and 93%, respectively, marking an improvement of up to 84% compared to baseline models not employing PowerPrint.