Sensors of Smart Devices in the Internet of Everything (IoE) Era: Big Opportunities and Massive Doubts

Table 4

Device fingerprint classification.

Work	Devices	Sensors	Scenario	Features	Algorithms	Results
[18]	20 Androids	Gyroscope, accelerometer, magnetometer, microphone, and vibrator	4 scenarios: (a) smartphone on a table with and without vibration and (a) smartphone held in the hand with and without vibration	Time-domain and frequency-domain features	Random forest and naive Bayes	Accelerometer accuracy higher than both sensors. With the combination of all sensors, the identification accuracy exceeds 90%
[191]	17 Androids and 17 IOS	Microphone, speakers, and accelerometer	Three scenarios (wooden desk, metal cabinet, and windowsill)	Frequency response and FFT value	Maximum likelihood estimation (MLE), simple Euclidean distance-based classification, and -NN classification	95% accuracy with a microphone and speaker and more than 98% for both
[180]	10 Androids	Accelerometer, gyroscope, magnetometer, microphone, camera, and vibrator	Flat wooden surface and hand-held	Photo response nonuniformity (PRNU), time-domain and frequency-domain features	Bagged decision tree	High accuracy for gyroscope and accelerometer, 100% for the combination of both
[192]	4 IOS, 1 Blackberry, and 8 Androids	Camera	—	Wavelet features, photo response nonuniformity (PRNU)	SVM	Accuracy of approximately 94%
[193]	8000 IOS	All sensors and context features	—	29 different features	SVM and random classifier	Accuracy of approximately 97%
[194]	6 cameras and 3 smartphones	Camera	—	Color, quality, and frequency-domain features	SVM	Accuracy between 66% and 97%
[195]	12 smartphones and camera	Camera	—	Color, quality, frequency domain, and wavelet feature+PRUN	SVM	For all features, accuracy increases. Some features obtain better results in specific scenarios
[196]	Arduino and accelerometer	Accelerometer	On a flat table	Time-domain features	Statistical	Each accelerometer chip has its own fingerprint
[25]	3 smartphones from three vendors	Accelerometer and gyroscope	On a flat table	Time-domain features	SVM	Accuracy more than 90%
[181]	30 between IOS and Android	Accelerometer and gyroscope	On a table	Time- and frequency-domain features	SVM, naive Bayes, multiclass decision tree, -nearest neighbor (KNN), quadratic discriminant Analysis (QDA) classifier and Bagged Decision Trees	Bagged decision trees have the highest accuracy