Detecting Fraud With Oversampling Techniques and Sparsity Contraints

Project Overview

Fraud detection is prevalent now more than ever due to the massive surge in the usage of online platforms. Many techniques exist to combat fraud; however, they often fail to capture the imbalanced class in data involving fraudulent activities. Our research contributes to the study of such concern with a model that harnesses the strengths of many existing models. We propose a solution that utilizes a combination of oversampling techniques and sparsity constraints to balance and predict fraud data.

Structure

Project
├── README.md
├── visualization_generation.py
├── requirements.txt
├── Baseline Models
    ├── GAT Baseline.py
    ├── GCN Baseline.py
    ├── GIN Baseline.py
    └── Fraud Sage-ROC.py
├── GraphSmote-main
    ├── data
    ├── data_load.py
    ├── main.py
    ├── models.py
    └── utils.py
└── GraphSmote-main-improved
    ├── data
    ├── data_load.py
    ├── main.py
    ├── models.py
    └── utils.py

Project Usage

1. Clone the repo to your local machine
2. Open your terminal and cd into the directory of the cloned repo
3. On the terminal, run pip install -r requirements.txt to obtain necessary packages to run the code.
4. Next, cd into the designated folder you want to run. And then run python (name of the py file).py. Explained more in detail below.

Baseline Models

Dependencies

• torch
• Numpy==1.21.5
• Pytorch==1.10.1
• DGL==0.8.2
• pygod == 1.1.0
• sklearn
• functools

Terminal Command to Run Code

run designated model

Ex: python GAT Baseline.py

GraphSMOTE

Terminal Command to Run Code

Run Pretrain

python main.py --imbalance --dataset=amazon --setting='recon'

Note: In checkpoint file, change "recon_300_False_0.5.pth" to "Pretrained.pth.pth", then run finetune

Run Finetune

python main.py --imbalance --dataset=amazon --setting='newG_cls' --load=Pretrained.pth

Improved GraphSmote

Terminal Command to Run Code

Run Pretrain

python main.py --imbalance --dataset=amazon --setting='recon'

Note: In checkpoint file, change recon_300_False_0.5.pth to Pretrained.pth.pth, then run finetune

Run Finetune

python main.py --imbalance --dataset=amazon --setting='newG_cls' --load=Pretrained.pth

Generate Visualizations

python visualization_generation.py