README.md

Introduction

This repository is credit scoring using data of Kaggle.
Competition name: Home Credit Default Risk
URL: https://www.kaggle.com/c/home-credit-default-risk

Requirement

• Python3
• scikit-learn
• LightGBM (https://github.com/Microsoft/LightGBM)
• Tensor Flow

Preprocessing & Feature Engineering

Preprocessing and feature engineering for each csv file is done four files in dataset directory.

• dataset/bureau.py
• dataset/credit_card_balance.py
• dataset/pos_cash_balance.py
• dataset/installments_payment.py

By executing dataset/make_dataset.py, the training data and test data subjected to the preprocessing
and feature engineering are output.

Train Model And Prediction

To train lightgbm model and mlp model.
To set files (train_X.npy, train_target.npy, test.npy) in all directory. There files created by previous step.
Execut below command.

python3 model_lightgbm.py
python3 model_mlp.py

Leared model save in all/tensor_model.
The prediction of each model is saved in all/out_gbm and all/out_mlp.

Submission

By executing src/model/submission.py, the submission file saved in all/submission.csv.