CNS_final_proj_EDOS_Detection_Fooling

Preprocess

python preprocess.py

• preprocess.py will generate two file, train and test, which are the inputs of libSVM.
• Details
  • For the row written in float or int, I did normalization.
  • For the row written in words, such as attack categories, I applied one hot to represent each category.

python preprocess_csv.py

• preprocess_csv.py generates two csv files, train.csv and test.csv under preprocess directory.
• Details
  • Map attack category to multiclass label
  • Drop rows containing NA (seems none)
  • Min-max normalization
  • One hot encoding

Grid Search to Reproduce Results

Analysis

The current implementation of analyze.py takes one or two .csv files in the format of UNSW_NB15 as input.

• For one file, comment the comparison section. Running analyze.py outputs analysis that contains the average and standard deviation of each continous data category.
• For two files, running analyze.py outputs analysis that contains the aforementioned information for both files, and also the difference between the two (both by value and by value / std).

Environment Issues

• Running preprocess.py requires the module ordered_set If running the following command

  pip install ordered_set

  throws the error

  AttributeError: module 'lib' has no attribute 'X509_V_FLAG_CB_ISSUER_CHECK'

  , simply edit the crypto.py file mentioned in the stacktrace and remove the offending line with #. Then, if

  pip install ordered_set

  throws the error

  AttributeError: module 'lib' has no attribute 'OpenSSL_add_all_algorithms'

  , uninstall pip with

  sudo apt remove python3-pip

  and run

  sudo python3 get-pip.py

  Reboot and run

  pip install pyopenssl --upgrade

  .Then, you should be able to install ordered_set and thus run preprocess.py

• Before training / grid search, run

  apt-get update

  and

  apt install libsvm-tools