TUM Logo

Machine Learning for Android Malware Detection in an Adversarial Environment

Machine Learning for Android Malware Detection in an Adversarial Environment

Supervisor(s): Bojan Kolosnjaji
Status: finished
Topic: Machine Learning Methods
Author: Donika Mirdita
Submission: 2019-03-29
Type of Thesis: Guided Research

Description

Mobile devices have become a ubiquitous presence and much of our social and economical life has moved on to their platforms.
Android devices make about 88\% of the global mobile market share while the platform itself is considerably vulnerable to malicious actors.
This situation requires for a deep dive into android malware detection strategies and efficient static analysis of malware code in order to enable the flagging of suspicious apps even prior to installation. Given the rise of adversarial attacks that enable subtle manipulations to bypass detectors, the analytic approaches need also to be checked for their robustness against these attacks. This guided research combines a set of approaches in order to develop a model for detecting malware through static analysis, and measures the robustness of the approach in an adversarial environment by using a modified version of the classical gradient attack.