Project 4: Credit Risk Assessment

Sunday, 02/26/2023

A project for finance industry or a hypothetical bank. Using machine learning models to predict whether a particular customer will default on their loan or not, based on different demographic indicators collected for the individual.

Project code:

Github Repository: credit-risk-assessment

Project Hosted on:

GCP App: Credit Risk Modeling App deployed on GCP

Importance:

Banks want to reduce their credit risk. The recession in 2008 in the US happened because mortgage loans were given to people with bad credit scores resulting in a lot of people defaulting due to which a lot of banks and investors had to face huge financial losses.

Glossary:

Credit Risk Modeling : Using data about a person to predict the probability of that person paying back the loan.
Default risk : A person’s probability to not pay back their loan. Banks use this to calculate the interest rate for a person. They charge people interest rates that are proportional to their default risk.

About the Dataset:

Data Source: Credit Risk Dataset, Kaggle

Column	Description
loan_status	Loan status	0 is non default 1 is default
person_age	Age	numerical
person_home_ownership	home ownership status	RENT, MORTGAGE, OWN, OTHER
person_emp_length	Employment length (in years)	numerical
loan_intent	Whether they own a home or not	PERSONAL, EDUCATION, MEDICAL, VENTURE, HOMEIMPROVEMENT, DEBTCONSOLIDATION
loan_grade	Loan grade	‘D’, ‘B’, ‘C’, ‘A’, ‘E’, ‘F’, ‘G’
loan_amnt	Loan amount	numerical
loan_int_rate	Interest rate	numerical
loan_percent_income	Loan to income ratio	numerical
cb_person_default_on_file	Historical default	‘Y’, ‘N’
cb_person_cred_hist_length	credit history length	numerical

Models and Technologies:

Azure Data Studio: Data in csv file loaded to SQL database created in Azure Data Studio. SQL server started on a Docker container.
Tableau: Connected to SQL database and built a dashboard for data visualization.

Python Jupyter Notebook: Built a machine learning pipeline using sklearn, pandas and XGboost to develop a classifier for predicting whether a person will default on their loan or not.
Flask Web application: Developed a python application using flask that takes certain inputs and used the already trained model in the backend to predict whether this new person will default on their loan or not.
Google Cloud Platform: Hosted the Flask app on GCP.