Cost-Aware Feature Selection for IoT Device Classification

The classification of Internet-of-Things (IoT) devices into different types is paramount importance, from multiple perspectives, including security and privacy aspects. Recent works have explored machine learning techniques for fingerprinting (or classifying) IoT devices, with promising results. However, the existing assumed that features used building models are readily available or can be easily extracted network traffic; in other words, they do not consider costs associated feature extraction. In this work, we take a more realistic approach, argue extraction has cost, features. We also step forward current practice considering misclassification loss as binary value, make case losses based on performance. Thereby, importantly, introduce notion risk device classification. define formulate problem cost-aware This being combinatorial optimization problem, develop novel algorithm to solve it fast effective way using cross-entropy (CE)-based stochastic technique. Using traffic real demonstrate capability CE-based selecting minimal while keeping cost within specified limit.