AMC-MDL: A Novel Approach of Android Malware Classification using Multimodel Deep Learning

2020 
With the continuous expansion of the Android operating system and the market for mobile apps continues to expand, the number of malwares designed for Android is also exploding. Therefore, identifying and detecting malware becomes a challenging task in mobile security. Because of the low detection efficiency by using traditional machine learning methods, we propose a novel Android malware detection method based on multi-model deep learning. Specifically, we select four features of permissions, API, Intent, and hardware components to build the feature vector for each app. In our study, we uses three different deep learning models: DNN (1st Model), Attention-CNN (2nd Model) and Attention-CNN-GRU (3rd Model). When the 1st model and the 2nd model are trained to achieve the optimal classification effect, the parameters of the optimal model training are selected as the input of the 3rd model. Finally, we evaluate the performance of our method on 5,560 malicious and 16,666 benign datasets. The experimental results show that the accuracy of the proposed detection method is 98.74%, which has an obvious competitive advantage over other methods.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    30
    References
    0
    Citations
    NaN
    KQI
    []