Abstract: In recent years,with the rapid development of mobile Internet,more and more businesses have moved from the browser to the mobile.But the black industry chain that is parasitic on the mobile Internet has reached the point of flooding.To solve this problem,the device fingerprint,that is,the use of the device’s characteristic attributes to generate a unique identifier for each device came into being.Many algorithms based on machine learning methods for device uniqueness authentication have emerged,most of which focus on the establishment of models.Few of them have in-depth research on feature selection.However,feature selection is directly related to the performance of the final model.Aiming at this problem,this paper proposes a new device fingerprint feature selection and model construction method (FSDS-WSC),which is based on the feature discrimination of different devices and the feature stability of the same device to select some of the most valuable features.The importance of the selected features’ weights is applied to the later model establishment.The FSFS-WSC is compared with other mainstream feature selection methods on 6424 Android devices in the real sence.The results show that FSFS-WSC has a great improvement compared with other methods,and the accuracy of device uniqueness authentication reaches 99.53%,which shows the superiority of FSFS-WSC.

Key words: Device fingerprint, Feature selection, Similarity, Weight, Discrimination, Stability