MaskDroid

Official code of "MaskDroid: Robust Android Malware Detection with Masked Graph Representations" published on 39th IEEE/ACM International Conference on Automated Software Engineering (ASE '24)

Overview

MaskDroid aims to build a powerful malware detector with remarkable robustness against adversarial attacks.  

Installation

Main packages: torch==2.3.1

Run DIR

To run effectiveness experiments:

python main.py --modeltype PreModel_v3 --batch_size 32 --lr 1e-3  --train_year all --need_pretrain --need_record --mask_rate 0.8

To run adversarial attack:

python model/attack.py --modeltype PreModel_v3 --white_box 

or for blackbox attack:

python model/attack.py --modeltype PreModel_v3

To run concept drift experiment:

python main.py --modeltype PreModel_v3 --concept_drift --sh --restore_epoch  19  --train_year 2019 --test_year 2020  --batch_size 32 --lr 1e-3  

Complete list of scripts are available in /scripts