RayminQAQ / MalDetection_image

Created in 2024/3/17, using pytorch and MOTIF dataset.

Malware Detection / Malware Classification

Purpose: Detect malware in MOTIF dataset using CNN.

To download MOTIF dataset: click here, or you an unzip MOTIF.7z in this poject.

About our repository

The project is run in Python 3.11.4 and cuda version 12.3, package dependencies are stored in requirement.txt.

You should download the MOTIF dataset first, then run the code as follow in below:

1. conversion_HW_v2.py:

   Turn binary executable into images

2. classification_HW/data_Inspection.py:

   Downscale some classes (init: lower than 40 instances) and turn images into ImageFolder acceptable format.

   • To get fimilair with ImageFolder, you can reference to Pytorch documentation

3. classification_HW/run.py:

   Train and test deep learning model.

   • dependency: classification_HW/model.py

   • dependency: classification_HW/image_save_data

     -> Our custom dataset. You can also use classification_HW/data_Inspection.py to create it on your own.

Model result

We achieved an accuracy of 80.25% on our custom dataset 'image_save_data'.

Credit

• Credit to RayminQAQ for leading the team and creating the entire Machine Learning pipeline, all stored in the '/classification_HW' folder.

• Credit to benson5104 for the excellent work on conversion_HW_v2.py.

• Credit to StanNTUST for providing computing resources and beautifying the coding style.