Loading...
Please wait, while we are loading the content...
Similar Documents
Paranoid Android : Android Malware Classification Using Supervised Learning on Call Graphs
| Content Provider | Semantic Scholar |
|---|---|
| Author | Mendoza, Mark Zhu, Michael |
| Copyright Year | 2017 |
| Abstract | Malware design and detection is an eternal arms race of increasing sophistication. A new front has been recently expanded in the discipline of malware obfuscation and self-modification, seeking to fool the signature-based approaches dominant in commercial anti-virus software. In response, security researchers have been seeking to design methods to classify executables based on their semantic function rather than their syntactic contents. The most interesting of these approaches has been to apply the powerful tools of network analysis to the graph defined by the state transitions of an application: its control flow graph. In this paper, we present our novel approach of Android malware classification based on supervised learning on the network properties of control flow graphs. Our program extracts the control flow of a given .apk using FlowDroid, then extracts a collection of network properties of the control flow graph using Snap.py, then retrieves an identification of that malware into a specific "family" from a XGBoost classifier, which we have trained on a labeled dataset. Using this approach, we were able to achieve upwards of 74% classification accuracy into specific families (different versions of a single exploit), and upwards of 98% classification accuracy into malware categories (e.g. Trojans, Adware, Ransomware). |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | http://web.stanford.edu/class/cs224w/projects/cs224w-19-final.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
