Fig 1.
The process of feature extraction.
Fig 2.
The form of feature vector.
Fig 3.
Feature classification.
Fig 4.
The system architecture of the proposed approach.
Fig 5.
Densely concatenated convolution.
Fig 6.
The architecture of the proposed model based on DenseNet.
Table 1.
The parameters of the constructed model based on DenseNet.
Table 2.
The details of the dataset.
Table 3.
The details of the extracted features.
Table 4.
The details of the features processed by FSMW.
Table 5.
The top ten permissions of the frequency similarity weight value.
Table 6.
The top ten intents of the frequency similarity weight value.
Table 7.
The top ten hardware of the frequency similarity weight value.
Table 8.
The top ten API calls of TF-IDF difference.
Table 9.
Some of the selected feature pairs with strong relationships.
Table 10.
The details of the features processed by CAPCC.
Table 11.
The details of the features processed by RFECV.
Table 12.
The details of the features processed by three levels of feature selection methods.
Table 13.
The influence of three levels of feature selection methods on the effect of feature set.
Table 14.
The comparison with machine learning algorithms.
Fig 7.
The ROC curve of our method.
Fig 8.
The AUCs of our method and machine learning algorithms.
Table 15.
The computational overhead and the number of parameters of each model.
Fig 9.
densenet_accuracy.
Fig 10.
densenet_loss.
Fig 11.
resnet_accuracy.
Fig 12.
resnet_loss.
Fig 13.
vgg16_accuracy.
Fig 14.
vgg16_loss.
Table 16.
The performance of each neural network.
Table 17.
The performance of each neural network.