mandyCheen/GraphTheoryDetector

IoT Malware Detection based on Graph Theory

Graph Theory Detector

ref. ALASMARY, Hisham, et al. Analyzing and detecting emerging Internet of Things malware: A graph-based approach. IEEE Internet of Things Journal, 2019, 6.5: 8977-8988.

Introduction

Description

• The main program is a malware detector.

  • Input : a binary file
  • Output : prob. of each class predicted by ML model
  • Flow :
    • reverse the bin and extract the feature
    • load the model
    • predict
    • print the result
    • write the result -> output csv file

Feature Extraction

• We reverse the binary file to function call graph(FCG) by r2pipe, then extract the attribute of FCG:

  • number of nodes

  • number of edges

  • density

  • (mean, max, min, median, std) of closeness_centrality

  • (mean, max, min, median, std) of betweenness_centrality

  • (mean, max, min, median, std) of degree_centrality

  • (mean, max, min, median, std) of shortestpaths.avglen

  The dimension of feature is 23.

Requirements

• python3
• radare2
• python package
  • r2pipe
  • networkx
  • joblib
  • argparse
  • sklearn

Files

• FeatureConstruction : some code about gpickle to feature.csv (for training, validation)
• Modeling : some code about training and saving the model
• MD(FC)_Model : save the model with .joblib, e.g. rf.joblib
• TestingBin : some binary for testing
• main.py : the detector(classifier)
• graphity.py : module for feature extraction
• utils.py : some useful func.
• scaler.joblib : scale the feature vector

Usage

• unzip models to FC_Model and MD_Model: make
• input binary: -i <path>, --input-path <path>
• model: -m <model>, --model <model>
  • rf, mlp, knn, svm
• output (record): -o <path>, --output-path <path>
• Malware Detection / Family Classification
  • do nothing if you wanna do malware detection(binary clf)
  • add -c if you wanna do family classification
• e.g. python main.py -i testingBin/0021eaf2 -o myDetector_FC_records.csv -m rf -c
  • using trained rf family classifier(-c), predict '0021eaf2' and write the result to 'myDetector_FC_records.csv'
  • add -W ignore if you keep getting bothered by warning msg

note

• a output csv file for a experiment

  • e.g.

    for bin in bins:
      cmd = 'python main.py -i ' + bin
      cmd += ' -m rf -o GraphTheoryDetector_RF_MD.csv'
      os.system(cmd)
    

    specify another output file if you wanna do FC task or you wanna change the model

• output file format

  Filename	Benignware	Malware
  00ffe391	0.97	0.03
  00f391fe	0.967	165.51
  1fe00f39	-1

  • it will record the prob of each class
  • -1 means fail

Model Performance

• Malware Detection

  Models	Accuracy	Time cost
  RandomForest	0.97	5.01
  KNN	0.967	165.51
  SVM	0.772	1606.21
  MLP	0.879	49.51

• Family Classification

  Models	Accuracy	Time cost
  RandomForest	0.952	4.21
  KNN	0.944	104.92
  SVM	0.71	1562.06
  MLP	0.887	70.99