HOME CREDIT DEFAULT RISK

GROUP NAME: DOM DOM

INSTRUCTOR: Ph.D DINH THUY TIEN

DO HOANG ANH

VUONG HOANG GIANG

A44498

A44526

CORE TEAM MEMBER

OVERVIEW

TARGET

3

2

1

Focus on customers aged 28 - 65, as this is the most common age range for loan borrowers.

AGE

CASH LOANS

Select only customers using cash loans, as this is the most common loan type.

EXISTING CUSTOMERS

Select only customers who have had a previous loan with Home Credit.

SCOPE

1, EXPLANATION

• What is Home Credit ?
• What is the main product ?
• How does Home Credit work?
• What is the mission?
• Business problem

2, DATA PREPARATION

• Overview Dataset
• Pre-processing data
• Create new features
• Feature selection

3, SOLUTION

• Model selection
• Prediction variables
• Choose the best cut-off threshold
• Result & Evaluation
• Advantages for business

​

​

​

4, APPLICATION

• Criteria
• Mini – App
• Review
• Expansion

AGENDA

EXPLANATION

WHAT IS HOME CREDIT ?

Home Credit is a financial institution specializing in offering loans to individuals with limited or restricted access to traditional banking services.

WHAT IS THE MAIN PRODUCT ?

​

POS CASH

INSTALLMENT

CREDIT CARD

HOW DOES HOME CREDIT WORK?

WHAT IS THE MISSION ?

Receive application

Review

&

Evaluation

Disburse

➤

➤

➤

Mitigating risk while maintaining business growth

Monitors

Reduce the rate of financial loss from customers who are facing difficulties

Build a model to predict if customers will struggle with loan repayment

Insufficient credit history for many customers makes accurate risk assessment challenging.

HIGH RISK

INCONSISTENT DATA

Information scattered across multiple platforms leads to difficulties in ensuring data uniformity and reliability.

TIME-CONSUMING

Fragmented and disorganized customer records across different sources slow down processes.

BUSINESS PROBLEM

DATA PREPARATION

OVERVIEW DATASET

MERGE DATA

​

APPLICATION_TRAIN

PREVIOUS_APPLICATION

• Rows: 38
• Columns: 264668

• Customer loans history at Home Credit
• Rows: 37
• Columns: 1670214

• Customer demographic and loan application data
• Rows: 122
• Columns: 307221

-

DATA TYPE

DUPLICATES

Removing duplicates improves dataset quality and model reliability.

MISSING VALUES

Imputation handles gaps in data to maintain accuracy and integrity.

PRE-PROCESSING

Ensuring correct data types (numerical/categorical) avoids errors in model training.

CREATE NEW FEARTURES

DAYS_BIRTH

PREVIOUS_APPLICATION

• Negative value ( - )
• Calculated by days

AGE

• Positive value ( + )
• Calculated by years

SUM

MEAN

• AMT_ANNUITY
• AMT_APPLICATION
• AMT_DOWN_PAYMENT
• AMT_GOODS_PRICE
• CNT_PAYMENT
• RATE_DOWN_PAYMENT

SOLUTION

MODEL SELECTION

High

Interpretability

Random Forest

K-Neareast Neighbors

(KNN)

Decision Tree

High

Poor

Poor

Logistic Regression

Implementability

PREDICTION MODEL

YES

NO

SELECTED FEARTURES

INPUT VARIABLES

LOGISTIC MODEL

OUTPUT

Having trouble paying debt?

MOST IMPORTANT FEATURES

SELECTED FEARTURES

FEATURES SELECTION

CHOOSE THE BEST THRESHOLD

Threshold: 0.162

Precision: 0.88

AUC: 0.703

BUSINESS VALUE

4776

62748

2310

864

=> Increase the ability to predict customers who are facing difficulty by 33%

ADVANTAGE

DISADVANTAGE

High Precision:The model is accurate when predicting difficulties

Low Recall: Indicating the model misses many actual cases of customers in difficulty

APPLICATION

EASY TO UNDERSTAND

HIGH APPLICABILITY

HIGH

PRECISION

EASY TO UPDATE

MINI APP

CRITERIA

https://s.pro.vn/FMyl

Implement functionality to convert relevant model outputs or inputs into string format for enhanced compatibility and data handling

CONVERT TO STRING

COMPLETE APP

Including all variables allows the model and using additional models to improving its accuracy and predictive power

BETTER INTERRFACE

Develop a more user-friendly interface to streamline interactions with the model, improving usability and accessibility

EXPANSION

REFERENCE SOURCE

• https://www.kaggle.com/code/kmanojus/classification-regression-problem
• https://www.kaggle.com/code/gpreda/home-credit-default-risk-extensive-eda/notebook
• https://www.kaggle.com/code/willkoehrsen/start-here-a-gentle-introduction
• https://www.kaggle.com/code/phamdinhkhanh/home-credit-default-risk
• https://www.kaggle.com/competitions/home-credit-default-risk/discussion/63032

THANK YOU