High-accuracy model recognition method of mobile device based on weighted feature similarity

Abstract

Abstract Accurately model recognition of mobile device is of great significance for identifying copycat device and protecting intellectual property rights. Although existing methods have realized high-accuracy recognition about device’s category and brand, the accuracy of model recognition still needs to be improved. For that, we propose Recognizer, a high-accuracy model recognition method of mobile device based on weighted feature similarity. We extract 20 features from the network traffic and physical attributes of device, and design feature similarity metric rules, and calculate inter-device similarity further. In addition, we propose feature importance evaluation strategies to assess the role of feature in recognition and determine the weight of each feature. Finally, based on all or part of 20 features, the similarity between the target device and known devices is calculated to recognize the brand and model of target device. Based on 587 models of mobile devices of 17 widely used brands such as Apple and Samsung, we carry out device recognition experiments. The results show that Recognizer can identify the device’s brand and model than existing methods more effectively. In average, the model recognition accuracy of Recognizer is 99.08% (+ 9.25%↑) when using 20 features and 92.08% (+ 29.26%↑) when using 13 features.