Journal
IEEE INTERNET OF THINGS JOURNAL
Volume 10, Issue 10, Pages 8560-8577Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/JIOT.2022.3194881
Keywords
Malware; Feature extraction; Internet of Things; Detectors; Static analysis; Deep learning; Data mining; Android malware; cyberattack; deep learning (DL); machine learning (ML); security
Ask authors/readers for more resources
Android-based mobile devices have become popular due to their ease of use and wide range of capabilities. However, this popularity has also attracted attackers who use sophisticated malware obfuscation and detection avoidance tactics. In this article, an improved deep neural network called AMDI-Droid is presented to safeguard Android devices from malicious apps. The model combines multiple hidden layers to learn effective feature representations and uses a blending approach to produce final predictions.
Android-based mobile devices have attracted a large number of users because they are easy to use and possess a wide range of capabilities. Because of its popularity, Android has become one of the most important platforms for attackers to launch their nefarious schemes. Due to the rising sophistication of Android malware obfuscation and detection avoidance tactics, many traditional malware detection approaches have become impractical due to their limited representation capabilities. Inspired by the success of deep learning in representation learning, this article presents an effective improved deep neural network to safeguard Android devices from malicious apps called AMDI-Droid. The presented approach contains three enhancements: 1) from the ensemble classifier perspective, we propose a new architecture based on a deep neural network, where the predictive outputs obtained from all hidden layers are blended to produce a final prediction; 2) the first hidden layer learns an effective feature representation from the original data through multiple subnetworks; and 3) a loss function is formulated by combining the predictive loss of each base classifier connected to the corresponding hidden layer. The superior performance of the proposed model is verified via intensive evaluations against state-of-the-art techniques in terms of the accuracy, precision, recall, F1-score, and Matthews correlation coefficient (MCC) metrics.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available