Table 1.

Comparison of the proposed work with state-of-the-art techniques.

Pap.	Purpose	Identification Method	Considered Features	Implementation	Attack Status
[20]	To identify the device type and device model	Calculating the similarity of features	Communication features extracted from header	Network cameras and factory-used devices	No attack
[21]	To employ behavioral fingerprinting for identification and authentication	K-nearest-neighbors (K-NN), Decision Trees (DT), gradient boosting, and majority voting	Header feature and payload-based features	14 home IoT devices	No attack
[4]	To automatically classify the IoT devices using TCP/IP packets	ML algorithms (DT, K48, OneR, PART) to classify device type	GA to determine most unique features from network, transport, and application layer	a database from [18]	No attack
[22]	To identify IoT devices using ML algorithms on network traffic data	Two-stages classifier: I. distinguish IoT vs non-IoT II. determine device class	features from network, transport, and application layer + data from Alexa Rank and GeoIP	9 distinct IoT devices, and PCs and smartphones	No attack
[23]	To identify IoT device types from the white list	multi-class classifier using RF	Features from Transmission Control Protocol/Internet Protocol (TCP/IP) sessions	17 different IoT devices (9 device type) by different vendors	Based on local organizational security policies violations
[24]	To classify IoT devices using traffic characteristics	multi-stage ML: Stage-0. Naïve Bayes Stage-1. RF	statistical attributes: activity cycles, port number, signaling patterns, and cipher suites	a living lab with 28 IoT devices	User Datagram Protocol (UDP) reflection and TCP SYN attacks
[26]	To recognize IoT devices by analyzing the generated network traffic	RF, DT, Support Vector Machine (SVM), k-NN, Artificial Neural Network and Gaussian Naive Bayes	Size of first 10 pack sent/ received and interval times	experimental smart home network of 4 devices	No attack
[25]	To automatically identify white-listed device types	ML classifiers ( e.g., SVM and K-NN)	behavioural and flow-based features	31 off-the-shelf IoT device (27 device types)	Adversaries compromising devices on network
[27]	To identify device-type without human intervention	unsupervised learning method	4 types of features: periodic flaws, periodic accuracy, period duration, and period stability	a dataset comprising 33 typical commercial IoT devices	Spoofing device fingerprints
Our work	To identify the device using device profiling	ML methods (RF, SVM, and Logistic Regression (LR))	header information, sensor measurements, and statistical features	2 types of sensors in an office	physical and remote attacks (Object emulation and Botnet attack)