1 of 19

PREDICTION�LENDING CLUB LOAN DATA

NGOI SAO HY VONG TEAM

2 of 19

A45070

Trương Đức Hoà

A44911

Dương Diệu Ly

CORE TEAM MEMBER

A44498

Đỗ Hoàng Anh

3 of 19

TARGET

MISSION

  • Predict ability to repay debt
  • Classify customer

STAKEHOLDERS

  • Lending Club
  • Investor

GOALS

  • Accuracy > 70%
  • AUC > 65%

1

2

3

4 of 19

AGENDA

2

DATA PREPARATION

  • Overview
  • Feature selection
  • Achieving data balance

​

​

SOLUTION

  • Model selection
  • Input – Output
  • AUC – Accuracy
  • Model application

3

IMPLEMENTATION

  • Mini app simulation.

​

4

1

EXPLANATION

  • What is Lending Club?
  • How does it work ?
  • Demand of Lending Club ?

​

​

​

5 of 19

EXPLANATION

6 of 19

WHAT IS LENDING CLUB?

HOW DOES IT WORK?

WHAT IS THE DEMAND OF LENDING CLUB?

A Peer-to-Peer lending intermediate platform

A place for borrowers and lenders can connect through an online system

Receive applications from borrowers

Lenders can decide to invest or not.

 Analyzes and evaluates them to lenders

Improve their prediction ability

Increase the prediction rate more accurately.

➤

➤

➤

7 of 19

DATA� PREPARATION

8 of 19

DATASET OVERVIEW

SCOPE

  • Over 765,000+ observation

INPUT

  • 12 variables

TIME

  • Collected from 2007 to 2018

01

02

03

9 of 19

Remove unnecessary columns, redundant columns (to avoid multicollinearity), & racial columns.��Consider statistically signifficant features only (Min– max scaling, Logistic Regression used)

FEATURE SELECTION

1,

2,

The amount of loan the customer wants to borrow

loan_amt

Loan term (months) the customer wants to pay

term

Interest Rate on the loan

Int_rate

The purpose of the customer borrowing money

purpose

Ratio of debt payments to total income

dti

Number of open credit lines in the borrower's credit file

open_acc

Total credit revolving balance

Revol_bal

Revolving line utilization rate

Revol_util

Number of mortgage accounts.

Mort_acc

Verification status of the borrower's income

Ver_status

Borrower's credit score

Fico

Number of derogatory public records

New_pub_rec

10 of 19

PROPORTION

➜

11 of 19

SOLUTION

12 of 19

MODEL SELECTION

13 of 19

INPUT VARIABLES

LOGISTIC MODEL

OUTPUT

loan_amnt

term

int_rate

dti

open_acc

revol_bal

​

​

​

​

​

revol_util

mort_acc

Ver_status

purpose

fico

new_pub_rec

​

Fully Paid

Charge Off

PREDICTION MODEL VARIABLES AND RESULTS

14 of 19

Charge Off

Fully Paid

Charge Off

Fully Paid

23783

40557

40557

222997

Prediction

Actual

CHOOSE THE BEST CUTOFF THRESHOLD

Threshold: 0,37

Accuracy: 0,75

AUC: 0,69

15 of 19

Without Model Case

Total Claims 327894

Fully Paid Claims 263554

Manual Failed 50% 131777

Total Failed 131777

With Model Case

Predict Positive 263554

True Fully Paid 222997

Manual Audit Fail 22% 49059

Model Fail Detection 40577

Total Failed 89636

ADVANTAGE

➜ Save 68% on guess wrong Fully Paid rate.

16 of 19

IMPLEMENTATION

17 of 19

01

EASY TO UNDERSTAND

HIGH APPLICABILITY

03

02

HIGH ACCURACY RATE

EASY TO UPDATE

04

MINI APP

Link: http://bit.ly/ngoi_sao_hy_vong

18 of 19

REFERENCE SOURCES

​

​

​

  • https://www.kaggle.com/code/emmaruyiyang/lending-club-loan-default-prediction-eda/notebook

​

​

19 of 19

THANKS FOR LISTENING