Performance Analysis of Extreme Learning Machine Classifiers on Radio Frequency Fingerprinting

Abstract

Internet of Things (IoT)is utilized in practically every industry. As IoT becomes more common, the number of wireless communication devices grows. The notion of security becomes more crucial as the number of devices and network grows.Due to welding constraints on IoT devices, the security can not be guaranteed.Radio frequency fingerprinting (RFF) methods, according to the literature, are utilized as an extra safety layer for wireless devices. Unique fingerprints due to the production defects of the devicesare used to identify wireless devices for security purposes in order to avoidfraud or fraud attempts. In this study, a ready-made dataset, consisting of 3985 registered samples and transformed to nine extracted features, from four WiFi Access Point(AP)devices was used. Using this data set, classification performances of Extreme Learning Machine (ELM), Constrained ELMs(CELMs), and Meta-ELM techniques are examined. Considering the classification performance of the Meta-ELM algorithm, it is concluded that it can be used in RF fingerprintingresearch due to itssuperior performance.The use of Meta-ELM in multiple classification problems will be a novelty in the literature.