Foreknow

This is a Pytorch implementation of the model in the paper "Forecasting Credit Default Risk with Graph Attention Networks".

Requirements

pytorch

pytorch-geometric

Usage

To run the code

1. Download the datasets at here.

2. Run the data processing code:

   python process.py

3. Run the credit default risk prediction code:

   python myGAT.py

Code Organization

1. process.py: This file is used to process the three raw datasets and output relevant attributes for future credit default risk prediction tasks.

• Input: "application_train.csv", "bureau.csv", "credit_card_balance.csv"
• Output: "one.csv"、"binary.csv", "r1_onehot.csv", "d1_onehot.csv", "l1_onehot.csv", "r1.csv", "l1.csv"

2. A&D_distance.Rmd: This file is used to calculate distances between mixed data by Ahmad & Dey method.

• Input: "r1.csv", "l1.csv"
• Output: "ahmad_r1.csv", "ahmadl1.csv"

3. myGAT.py: This file is the core implementation of the prediction model.

• Input: "r1_onehot.csv", "d1_onehot.csv", "l1_onehot.csv", "ahmad_r1.csv", "ahmadl1.csv", "binary.csv"
• Output: "pre.csv"