Insider Threat Graph Models (R5.2 & R6.2)

This repository contains data preparation, feature extraction, and graph-based models (GCN-NN, GCN-BiLSTM, and GCN-BiLSTM+Attention for implicit+explicit signals) for insider-threat detection using the CMU CERT Insider Threat Test Dataset (R5.2, R6.2).

Repository Structure

.
│   environment.yml
│   requirements.txt
│   tree.txt
│
├── data/
│   ├── original/
│   │   ├── r5_data/
│   │   └── r6_data/
│   ├── preprocessed/
│   │   ├── r5_data/
│   │   └── r6_data/
│   ├── r5.2-used-model/
│   └── r6.2_used_model/
│
├── notebook/
│   │   01_feature_extraction.ipynb
│   │   02_GCN-LSTM.ipynb
│   │   03_GCN.ipynb
│   │   04_GCN_Implicit_Explicit.ipynb
│   │   r5_data_preparation.ipynb
│   │   r6_data_preparation.ipynb
│   │   email_http_divider.ipynb
│
└── script/
    │   get_feature.py
    │   process_feature.py
    └── __pycache__/

The full, deep tree (including all .pkl and model checkpoints) can be kept in tree.txt to keep this README short.

Setup

1. Clone the repository

   root@kitploit:~

   git clone https://github.com/your-username/your-repo.git
   cd your-repo
   

2. Create the Conda environment

   root@kitploit:~

   conda env create -f environment.yml
   

3. Activate it

   root@kitploit:~

   conda activate <env_name>
   

   Replace <env_name> with the value under name: in environment.yml.

4. (Optional) Pip-only install

   root@kitploit:~

   pip install -r requirements.txt
   

Data

1. Download the CMU CERT Insider Threat Test Dataset (R5.2 and R6.2) from:
   https://kilthub.cmu.edu/articles/dataset/Insider_Threat_Test_Dataset/12841247

2. Place raw data under:

   root@kitploit:~

   data/original/r5_data/
   data/original/r6_data/
   

Preprocessing

Run the preparation notebooks to create the smaller session files and canonical CSVs/PKLs.

• R5.2

  root@kitploit:~

  notebook/r5_data_preparation.ipynb
  

• R6.2

  root@kitploit:~

  notebook/r6_data_preparation.ipynb
  

Expected R5.2 outputs (examples):

r5_small_device_session.pkl
r5_small_file_session.pkl
r5_small_email_session.pkl
r5_small_http_session.pkl
r5_small_sessoion_data.pkl
r5_small_session_log_df.pkl

Expected R6.2 outputs include:

r6_2_small_device_session.pkl
r6_2_small_file_session.pkl
r6_2_small_email_session.pkl
r6_2_small_http_session.pkl
r6_2_small_session_log_df.pkl
r6_2_user_df_pc.csv

Outputs are saved under:

data/preprocessed/r5_data/
data/preprocessed/r6_data/

Feature Extraction

Run:

notebook/01_feature_extraction.ipynb

Outputs:

log_activity_code.pkl
device_activity_code.pkl
file_activity_code.pkl
email_activity_code.pkl
http_activity_code.pkl

This notebook also generates graph lists/masks, e.g.:

data/preprocessed/r6_data/r6_graph_list_feature_corrected_range_corrected.pkl

Models

• GCN-BiLSTM (Explicit)

  root@kitploit:~

  notebook/02_GCN-LSTM.ipynb
  

• GCN-NN (Explicit)

  root@kitploit:~

  notebook/03_GCN.ipynb
  

• GCN-BiLSTM + Attention (Implicit + Explicit)

  root@kitploit:~

  notebook/04_GCN_Implicit_Explicit.ipynb
  

Pretrained checkpoints and ROC artifacts are under:

data/r5.2-used-model/
data/r6.2_used_model/

Each subfolder contains:

*.pth                 # model weights
roc_curve_data_*.npz  # saved ROC curve data
roc_curves*.png       # ROC plots