dianaamfr / loan-fraud-prediction

To Loan or not to Loan

A data mining project to help bank managers avoid trusting non-compliant clients with loans by predicting if a loan will end successfully based on data about the clients and previous loans. This project was developed during the Machine Learning course at FEUP.

Compilation

Database

From the src folder of the repository:

Create MySQL database:

1. mysql -u root -p

    1. CREATE DATABASE bank_database;
    2. SET GLOBAL local_infile = true;
    3. quit;

2. mysql -u root -p --local-infile=1 bank_database < database/database.sql

Graphviz

Also, to plot the trees you must install graphviz in your system.

https://graphviz.org/download/

Create the virtual environment

Ubuntu

1. python3 -m venv env
2. source env/bin/activate
3. pip3 install -r ../requirements.txt

Windows

1. py -m venv env
2. .\env\Scripts\activate.bat
3. pip install -r ..\requirements.txt

---

Run

1. Clean: Generate train and test csvs with clean data and save them to clean_data folder

make clean <submission_name>

• outputs clean_data/<submission_name>.csv
• e.g. make clean sub2 will generate the file sub2-train.csv and sub2-test.csv in the clean_data folder

2. Train: Train the model with the clean data, using a specific classifier, compute the AUC and store the model in the models folder

make train <classifier> <submission_name>

• outputs models/<classifier>-<submission_name>.sav
• e.g. make train logistic_regression sub2 will use as input the file sub2-train.csv from the clean_data folder and store in the models folder the model that results of applying the Logistic Regression Classifier to the data - logistic_regression-sub2.sav

3. Test: Test a model with the test data and store the result in the results folder

make test <classifier> <submission_name>

• outputs results/<classifier >-<submission_name >.csv
• e.g. make test logistic_regression sub2 will apply the model models/logistic_regression-sub2.sav to the data from clean_data/sub2-test.csv and store in results/logistic_regression-sub2.csv

4. Explore: Explore the various datasets by printing some statistics and generating some plots

make explore <table>

• outputs generated plots in the folder data_understanding/plots
• Available tables: account, card, client, disp, district, loan, trans
• e.g. make explore account will perform data exploration to the table Account, saving some plots in the folders data_understanding/plots/distribution/account and data_understanding/plots/correlation/account

5. Clustering: Solve the descriptive problem, by generating some graphs describing the cluster approach to distinguish between different client types

make clustering

• outputs generated graphs that are opened in the browser

6. Clean Models: Empty the folder models containing the trained models

make clean_models

7. Clean Cache: Empty the Python cache folders (__pycache__)

Collaborators

1. Diana Freitas
2. Mariana Ramos
3. Paulo Ribeiro

loan-fraud-prediction's Issues

Clustering

1. só clientes (perfis muito individuais)
2. juntar distritos (perfis socio demograficos)
3. juntar transações e cenas financeiras (perfis de comportamento financeiro)

2 Clusters
Bonus: usar clusters para criar o perfil do cliente

Improve Data Understanding Presentation

Add subtitle and clear description

Add Tree Visualization

New Feature Selection methods