WQD7004 - Programming for Data Science
Predictive Modelling for Loan Approval (Group 13)
| Name | Matric No |
|---|---|
| Wan Nurul Adibah Binti Wan Tarmizi | 24062857 |
| Nur Anis Insyirah Binti Jowanis | 24073722 |
| Nur Izzati Binti Juhari | 24075056 |
| Puteri Safa Balqis Binti Megat Sharizal Amri | 23114952 |
| Yuhan Silvian | 24077249 |
Introduction
Financial risk assessment is crucial in loan approval decisions, enabling banks and credit institutions to mitigate defaults while ensuring credit access for qualified applicants. In an era where credit accessibility directly influences economic growth, the ability to accurately assess borrower risk has become both a strategic and regulatory necessity.
Traditional methods rely on manual evaluation of applicant profiles such as income, credit history, and employment type, but these are often time-consuming and subject to human bias (Debabrata Dansana et al., 2023; Wu, 2024).
Recent advances in machine learning, particularly Logistic Regression, have shown high potential in predicting default risk by analyzing relationships between multiple variables (N. Penchalaiah, 2022). However, model outcomes may still be influenced by organizational policies and subjective decision making in borderline cases (Jansson et al., 2023).
This project explores how machine learning can be used to increase accuracy and promote fairness in lending decisions by reducing reliance on subjective judgment.
Problem Statement
Financial institutions’ loan approval procedures are still beset by structural inefficiencies and antiquated risk assessment techniques that fall short in addressing today’s lending issues. Despite being widely used, traditional credit scoring models show serious limits in their capacity to assess non-traditional borrowers where the models might wrongly rejected qualified borrowers while approving high-risk candidates resulting in financial loss out of total loan portfolios annually.
Thus, there is an urgent need for a machine learning framework that maintains regulatory compliance and operational scalability for lenders while extracting useful insights from multidimensional applicant data.
Objectives
RO1: To preprocess the dataset and uncover patterns, trends, and relationships within the data, which will provide valuable insights and aid subsequent modeling or decision-making processes
RO2: To develop a binary classification model (approved/not approved) and regression model (RiskScore) using machine learning that integrates dynamic feature weighting to achieve more than 80% accuracy
RO3: To identify the key variables that influence the likelihood approved/not approved decision
RO4: To evaluate and compare between machine learning models in order to find the best-performing model
Research Questions
RQ1: How can a binary classification model be developed to accurately predict loan approval decisions?
RQ2: How can a regression model be constructed to predict the risk score of loan applicants?
RQ3: What are the most significant factors influencing loan approval outcomes?
RQ4: Which machine learning model demonstrates the highest performance in predicting loan approval and assessing default risk?
Chosen Dataset
Name: Financial Risk for Loan Approval
Description: This is a synthetic dataset containing 20,000 records of individual loan applicants, covering both personal and financial attributes
Link: https://www.kaggle.com/datasets/lorenzozoppelletto/financial-risk-for-loan-approval
Import Dataset
Install packages here:
options(repos = c(CRAN = "https://cloud.r-project.org"))
install.packages("caret")## Installing package into 'C:/Users/puter/AppData/Local/R/win-library/4.4'
## (as 'lib' is unspecified)## package 'caret' successfully unpacked and MD5 sums checked
##
## The downloaded binary packages are in
## C:\Users\puter\AppData\Local\Temp\RtmpEr2lRG\downloaded_packagesdata <- read.csv("Loan.csv")
head(data)## ApplicationDate Age AnnualIncome CreditScore EmploymentStatus EducationLevel
## 1 2018-01-01 45 39948 617 Employed Master
## 2 2018-01-02 38 39709 628 Employed Associate
## 3 2018-01-03 47 40724 570 Employed Bachelor
## 4 2018-01-04 58 69084 545 Employed High School
## 5 2018-01-05 37 103264 594 Employed Associate
## 6 2018-01-06 37 178310 626 Self-Employed Master
## Experience LoanAmount LoanDuration MaritalStatus NumberOfDependents
## 1 22 13152 48 Married 2
## 2 15 26045 48 Single 1
## 3 26 17627 36 Married 2
## 4 34 37898 96 Single 1
## 5 17 9184 36 Married 1
## 6 16 15433 72 Married 0
## HomeOwnershipStatus MonthlyDebtPayments CreditCardUtilizationRate
## 1 Own 183 0.35441792
## 2 Mortgage 496 0.08782697
## 3 Rent 902 0.13741410
## 4 Mortgage 755 0.26758714
## 5 Mortgage 274 0.32053532
## 6 Rent 732 0.10221134
## NumberOfOpenCreditLines NumberOfCreditInquiries DebtToIncomeRatio
## 1 1 2 0.35833560
## 2 5 3 0.33027367
## 3 2 0 0.24472911
## 4 2 1 0.43624427
## 5 0 0 0.07888421
## 6 5 1 0.25936640
## BankruptcyHistory LoanPurpose PreviousLoanDefaults PaymentHistory
## 1 0 Home 0 29
## 2 0 Debt Consolidation 0 21
## 3 0 Education 0 20
## 4 0 Home 0 27
## 5 0 Debt Consolidation 0 26
## 6 0 Debt Consolidation 1 16
## LengthOfCreditHistory SavingsAccountBalance CheckingAccountBalance
## 1 9 7632 1202
## 2 9 4627 3460
## 3 22 886 895
## 4 10 1675 1217
## 5 27 1555 4981
## 6 19 2118 1223
## TotalAssets TotalLiabilities MonthlyIncome UtilityBillsPaymentHistory
## 1 146111 19183 3329.000 0.7249720
## 2 53204 9595 3309.083 0.9351321
## 3 25176 128874 3393.667 0.8722406
## 4 104822 5370 5757.000 0.8961547
## 5 244305 17286 8605.333 0.9413687
## 6 67914 40843 14859.167 0.7560794
## JobTenure NetWorth BaseInterestRate InterestRate MonthlyLoanPayment
## 1 11 126928 0.199652 0.2275896 419.8060
## 2 3 43609 0.207045 0.2010771 794.0542
## 3 6 5205 0.217627 0.2125480 666.4067
## 4 5 99452 0.300398 0.3009108 1047.5070
## 5 5 227019 0.197184 0.1759902 330.1791
## 6 5 27071 0.217433 0.2176012 385.5771
## TotalDebtToIncomeRatio LoanApproved RiskScore
## 1 0.18107720 0 49
## 2 0.38985245 0 52
## 3 0.46215697 0 52
## 4 0.31309831 0 54
## 5 0.07020985 1 36
## 6 0.07521129 1 44Load Libraries:
install.packages(c("tidyverse","ggplot2", "dplyr", "skimr", "corrplot", "knitr"))## Installing packages into 'C:/Users/puter/AppData/Local/R/win-library/4.4'
## (as 'lib' is unspecified)## package 'tidyverse' successfully unpacked and MD5 sums checked
## package 'ggplot2' successfully unpacked and MD5 sums checked
## package 'dplyr' successfully unpacked and MD5 sums checked
## package 'skimr' successfully unpacked and MD5 sums checked
## package 'corrplot' successfully unpacked and MD5 sums checked
## package 'knitr' successfully unpacked and MD5 sums checked
##
## The downloaded binary packages are in
## C:\Users\puter\AppData\Local\Temp\RtmpEr2lRG\downloaded_packageslibrary(ggplot2)
library(dplyr)##
## Attaching package: 'dplyr'## The following objects are masked from 'package:stats':
##
## filter, lag## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, unionlibrary(skimr)
library(corrplot)## corrplot 0.95 loadedlibrary(tidyverse)## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ forcats 1.0.0 ✔ stringr 1.5.1
## ✔ lubridate 1.9.4 ✔ tibble 3.2.1
## ✔ purrr 1.0.4 ✔ tidyr 1.3.1
## ✔ readr 2.1.5## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errorslibrary(knitr)Data Overview
- Structure of data
- Summary of data
- Data quality check
# Structure of the dataset
str(data)## 'data.frame': 20000 obs. of 36 variables:
## $ ApplicationDate : chr "2018-01-01" "2018-01-02" "2018-01-03" "2018-01-04" ...
## $ Age : int 45 38 47 58 37 37 58 49 34 46 ...
## $ AnnualIncome : int 39948 39709 40724 69084 103264 178310 51250 97345 116841 40615 ...
## $ CreditScore : int 617 628 570 545 594 626 564 516 603 612 ...
## $ EmploymentStatus : chr "Employed" "Employed" "Employed" "Employed" ...
## $ EducationLevel : chr "Master" "Associate" "Bachelor" "High School" ...
## $ Experience : int 22 15 26 34 17 16 39 23 12 19 ...
## $ LoanAmount : int 13152 26045 17627 37898 9184 15433 12741 19634 55353 25443 ...
## $ LoanDuration : int 48 48 36 96 36 72 48 12 60 12 ...
## $ MaritalStatus : chr "Married" "Single" "Married" "Single" ...
## $ NumberOfDependents : int 2 1 2 1 1 0 0 5 5 4 ...
## $ HomeOwnershipStatus : chr "Own" "Mortgage" "Rent" "Mortgage" ...
## $ MonthlyDebtPayments : int 183 496 902 755 274 732 337 288 638 704 ...
## $ CreditCardUtilizationRate : num 0.3544 0.0878 0.1374 0.2676 0.3205 ...
## $ NumberOfOpenCreditLines : int 1 5 2 2 0 5 6 5 3 3 ...
## $ NumberOfCreditInquiries : int 2 3 0 1 0 1 1 0 0 2 ...
## $ DebtToIncomeRatio : num 0.3583 0.3303 0.2447 0.4362 0.0789 ...
## $ BankruptcyHistory : int 0 0 0 0 0 0 0 0 1 0 ...
## $ LoanPurpose : chr "Home" "Debt Consolidation" "Education" "Home" ...
## $ PreviousLoanDefaults : int 0 0 0 0 0 1 0 0 0 0 ...
## $ PaymentHistory : int 29 21 20 27 26 16 21 19 25 23 ...
## $ LengthOfCreditHistory : int 9 9 22 10 27 19 18 11 29 10 ...
## $ SavingsAccountBalance : int 7632 4627 886 1675 1555 2118 5161 781 1157 1028 ...
## $ CheckingAccountBalance : int 1202 3460 895 1217 4981 1223 1735 74 708 446 ...
## $ TotalAssets : int 146111 53204 25176 104822 244305 67914 65624 50177 29632 129664 ...
## $ TotalLiabilities : int 19183 9595 128874 5370 17286 40843 43894 11556 49940 12852 ...
## $ MonthlyIncome : num 3329 3309 3394 5757 8605 ...
## $ UtilityBillsPaymentHistory: num 0.725 0.935 0.872 0.896 0.941 ...
## $ JobTenure : int 11 3 6 5 5 5 5 5 3 3 ...
## $ NetWorth : int 126928 43609 5205 99452 227019 27071 21730 38621 7711 116812 ...
## $ BaseInterestRate : num 0.2 0.207 0.218 0.3 0.197 ...
## $ InterestRate : num 0.228 0.201 0.213 0.301 0.176 ...
## $ MonthlyLoanPayment : num 420 794 666 1048 330 ...
## $ TotalDebtToIncomeRatio : num 0.1811 0.3899 0.4622 0.3131 0.0702 ...
## $ LoanApproved : int 0 0 0 0 1 1 0 1 0 0 ...
## $ RiskScore : num 49 52 52 54 36 44 50 42.4 61 53 ...# Summary statistics
summary(data)## ApplicationDate Age AnnualIncome CreditScore
## Length:20000 Min. :18.00 Min. : 15000 Min. :343.0
## Class :character 1st Qu.:32.00 1st Qu.: 31679 1st Qu.:540.0
## Mode :character Median :40.00 Median : 48566 Median :578.0
## Mean :39.75 Mean : 59162 Mean :571.6
## 3rd Qu.:48.00 3rd Qu.: 74391 3rd Qu.:609.0
## Max. :80.00 Max. :485341 Max. :712.0
## EmploymentStatus EducationLevel Experience LoanAmount
## Length:20000 Length:20000 Min. : 0.00 Min. : 3674
## Class :character Class :character 1st Qu.: 9.00 1st Qu.: 15575
## Mode :character Mode :character Median :17.00 Median : 21915
## Mean :17.52 Mean : 24883
## 3rd Qu.:25.00 3rd Qu.: 30835
## Max. :61.00 Max. :184732
## LoanDuration MaritalStatus NumberOfDependents HomeOwnershipStatus
## Min. : 12.00 Length:20000 Min. :0.000 Length:20000
## 1st Qu.: 36.00 Class :character 1st Qu.:0.000 Class :character
## Median : 48.00 Mode :character Median :1.000 Mode :character
## Mean : 54.06 Mean :1.517
## 3rd Qu.: 72.00 3rd Qu.:2.000
## Max. :120.00 Max. :5.000
## MonthlyDebtPayments CreditCardUtilizationRate NumberOfOpenCreditLines
## Min. : 50.0 Min. :0.0009745 Min. : 0.000
## 1st Qu.: 286.0 1st Qu.:0.1607936 1st Qu.: 2.000
## Median : 402.0 Median :0.2666726 Median : 3.000
## Mean : 454.3 Mean :0.2863810 Mean : 3.023
## 3rd Qu.: 564.0 3rd Qu.:0.3906339 3rd Qu.: 4.000
## Max. :2919.0 Max. :0.9173801 Max. :13.000
## NumberOfCreditInquiries DebtToIncomeRatio BankruptcyHistory LoanPurpose
## Min. :0.000 Min. :0.00172 Min. :0.0000 Length:20000
## 1st Qu.:0.000 1st Qu.:0.16103 1st Qu.:0.0000 Class :character
## Median :1.000 Median :0.26445 Median :0.0000 Mode :character
## Mean :0.993 Mean :0.28573 Mean :0.0524
## 3rd Qu.:2.000 3rd Qu.:0.39033 3rd Qu.:0.0000
## Max. :7.000 Max. :0.90225 Max. :1.0000
## PreviousLoanDefaults PaymentHistory LengthOfCreditHistory
## Min. :0.0000 Min. : 8.00 Min. : 1.00
## 1st Qu.:0.0000 1st Qu.:21.00 1st Qu.: 8.00
## Median :0.0000 Median :24.00 Median :15.00
## Mean :0.1000 Mean :23.99 Mean :14.96
## 3rd Qu.:0.0000 3rd Qu.:27.00 3rd Qu.:22.00
## Max. :1.0000 Max. :45.00 Max. :29.00
## SavingsAccountBalance CheckingAccountBalance TotalAssets
## Min. : 73 Min. : 24 Min. : 2098
## 1st Qu.: 1542 1st Qu.: 551 1st Qu.: 31180
## Median : 2986 Median : 1116 Median : 60699
## Mean : 4946 Mean : 1783 Mean : 96964
## 3rd Qu.: 5873 3rd Qu.: 2126 3rd Qu.: 117405
## Max. :200089 Max. :52572 Max. :2619627
## TotalLiabilities MonthlyIncome UtilityBillsPaymentHistory JobTenure
## Min. : 372 Min. : 1250 Min. :0.2592 Min. : 0.000
## 1st Qu.: 11197 1st Qu.: 2630 1st Qu.:0.7274 1st Qu.: 3.000
## Median : 22203 Median : 4035 Median :0.8210 Median : 5.000
## Mean : 36252 Mean : 4892 Mean :0.7999 Mean : 5.003
## 3rd Qu.: 43147 3rd Qu.: 6163 3rd Qu.:0.8923 3rd Qu.: 6.000
## Max. :1417302 Max. :25000 Max. :0.9994 Max. :16.000
## NetWorth BaseInterestRate InterestRate MonthlyLoanPayment
## Min. : 1000 Min. :0.1301 Min. :0.1133 Min. : 97.03
## 1st Qu.: 8735 1st Qu.:0.2139 1st Qu.:0.2091 1st Qu.: 493.76
## Median : 32856 Median :0.2362 Median :0.2354 Median : 728.51
## Mean : 72294 Mean :0.2391 Mean :0.2391 Mean : 911.61
## 3rd Qu.: 88826 3rd Qu.:0.2615 3rd Qu.:0.2655 3rd Qu.: 1112.77
## Max. :2603208 Max. :0.4050 Max. :0.4468 Max. :10892.63
## TotalDebtToIncomeRatio LoanApproved RiskScore
## Min. :0.01604 Min. :0.000 Min. :28.80
## 1st Qu.:0.17969 1st Qu.:0.000 1st Qu.:46.00
## Median :0.30271 Median :0.000 Median :52.00
## Mean :0.40218 Mean :0.239 Mean :50.77
## 3rd Qu.:0.50921 3rd Qu.:0.000 3rd Qu.:56.00
## Max. :4.64766 Max. :1.000 Max. :84.00# Data quality check
dim(data)## [1] 20000 36# Get number of rows and columns
any(is.na(data))## [1] FALSE# Get number of missing values for each columns
kable(colSums(is.na(data)))| x | |
|---|---|
| ApplicationDate | 0 |
| Age | 0 |
| AnnualIncome | 0 |
| CreditScore | 0 |
| EmploymentStatus | 0 |
| EducationLevel | 0 |
| Experience | 0 |
| LoanAmount | 0 |
| LoanDuration | 0 |
| MaritalStatus | 0 |
| NumberOfDependents | 0 |
| HomeOwnershipStatus | 0 |
| MonthlyDebtPayments | 0 |
| CreditCardUtilizationRate | 0 |
| NumberOfOpenCreditLines | 0 |
| NumberOfCreditInquiries | 0 |
| DebtToIncomeRatio | 0 |
| BankruptcyHistory | 0 |
| LoanPurpose | 0 |
| PreviousLoanDefaults | 0 |
| PaymentHistory | 0 |
| LengthOfCreditHistory | 0 |
| SavingsAccountBalance | 0 |
| CheckingAccountBalance | 0 |
| TotalAssets | 0 |
| TotalLiabilities | 0 |
| MonthlyIncome | 0 |
| UtilityBillsPaymentHistory | 0 |
| JobTenure | 0 |
| NetWorth | 0 |
| BaseInterestRate | 0 |
| InterestRate | 0 |
| MonthlyLoanPayment | 0 |
| TotalDebtToIncomeRatio | 0 |
| LoanApproved | 0 |
| RiskScore | 0 |
Data Pre-Processing/Cleaning
The dataset includes the following columns:
1.
ApplicationDate: Loan application date *
ApplicationDate to be dropped and used as an account ID since it is
generated as one day one account
2. Age: Applicant’s age
3.
AnnualIncome: Yearly income
4.
CreditScore: Creditworthiness score * Mirrors
real-world credit scoring systems like FICO, where: * 300–579: Poor *
580–669: Fair * 670–739: Good * 740–799: Very Good * 800–850:
Excellent
5. EmploymentStatus: Job situation
6.
EducationLevel: Highest education attained
7.
Experience: Work experience in Years * Rename
Experience to WorkExperience
8. LoanAmount: Requested loan size
9.
LoanDuration: Loan repayment period in years
10.
MaritalStatus: Applicant’s marital state
11.
NumberOfDependents: Number of dependents
12.
HomeOwnershipStatus: Homeownership type
13.
MonthlyDebtPayments: Monthly debt obligations
14.
CreditCardUtilizationRate: Credit card usage percentage
15. NumberOfOpenCreditLines: Active credit lines
16. NumberOfCreditInquiries: Credit checks count
17. DebtToIncomeRatio: Debt to income proportion
18. BankruptcyHistory: Bankruptcy records
19.
LoanPurpose: Reason for loan
20.
PreviousLoanDefaults: Prior loan defaults
21.
PaymentHistory: Past payment behavior
22.
LengthOfCreditHistory: Credit history duration
23.
SavingsAccountBalance: Savings account amount
24.
CheckingAccountBalance: Checking account funds
25.
TotalAssets: Total owned assets
26.
TotalLiabilities: Total owed debts
27.
MonthlyIncome: Income per month
28.
UtilityBillsPaymentHistory: Utility payment record
29. JobTenure: Job duration
30.
NetWorth: Total financial worth
31.
BaseInterestRate: Starting interest rate
32.
InterestRate: Applied interest rate
33.
MonthlyLoanPayment: Monthly loan payment
34.
TotalDebtToIncomeRatio: Total debt against income
35. LoanApproved: Loan approval status
36.
RiskScore: Risk assessment score * A higher Risk Score
indicates a more significant risk.
# 1. Distribution of Loan Approved
ggplot(data, aes(x = factor(LoanApproved))) +
geom_bar() +
geom_text(stat = "count", aes(label = ..count..), vjust = -0.5) +
labs(title = "LoanApproved Distribution",
x = "LoanApproved",
y = "Count") +
theme(plot.title = element_text(hjust = 0.5))## Warning: The dot-dot notation (`..count..`) was deprecated in ggplot2 3.4.0.
## ℹ Please use `after_stat(count)` instead.
## This warning is displayed once every 8 hours.
## Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
## generated.
#Data Cleaning
#1. Drop applicationDate and changed it to AppID as Application ID
#2. Create another variable CreditScoreGroup to follow FICO
data$AppID <- gsub("-", "", data$ApplicationDate)
data$ApplicationDate <- NULL
head(data)## Age AnnualIncome CreditScore EmploymentStatus EducationLevel Experience
## 1 45 39948 617 Employed Master 22
## 2 38 39709 628 Employed Associate 15
## 3 47 40724 570 Employed Bachelor 26
## 4 58 69084 545 Employed High School 34
## 5 37 103264 594 Employed Associate 17
## 6 37 178310 626 Self-Employed Master 16
## LoanAmount LoanDuration MaritalStatus NumberOfDependents HomeOwnershipStatus
## 1 13152 48 Married 2 Own
## 2 26045 48 Single 1 Mortgage
## 3 17627 36 Married 2 Rent
## 4 37898 96 Single 1 Mortgage
## 5 9184 36 Married 1 Mortgage
## 6 15433 72 Married 0 Rent
## MonthlyDebtPayments CreditCardUtilizationRate NumberOfOpenCreditLines
## 1 183 0.35441792 1
## 2 496 0.08782697 5
## 3 902 0.13741410 2
## 4 755 0.26758714 2
## 5 274 0.32053532 0
## 6 732 0.10221134 5
## NumberOfCreditInquiries DebtToIncomeRatio BankruptcyHistory
## 1 2 0.35833560 0
## 2 3 0.33027367 0
## 3 0 0.24472911 0
## 4 1 0.43624427 0
## 5 0 0.07888421 0
## 6 1 0.25936640 0
## LoanPurpose PreviousLoanDefaults PaymentHistory LengthOfCreditHistory
## 1 Home 0 29 9
## 2 Debt Consolidation 0 21 9
## 3 Education 0 20 22
## 4 Home 0 27 10
## 5 Debt Consolidation 0 26 27
## 6 Debt Consolidation 1 16 19
## SavingsAccountBalance CheckingAccountBalance TotalAssets TotalLiabilities
## 1 7632 1202 146111 19183
## 2 4627 3460 53204 9595
## 3 886 895 25176 128874
## 4 1675 1217 104822 5370
## 5 1555 4981 244305 17286
## 6 2118 1223 67914 40843
## MonthlyIncome UtilityBillsPaymentHistory JobTenure NetWorth BaseInterestRate
## 1 3329.000 0.7249720 11 126928 0.199652
## 2 3309.083 0.9351321 3 43609 0.207045
## 3 3393.667 0.8722406 6 5205 0.217627
## 4 5757.000 0.8961547 5 99452 0.300398
## 5 8605.333 0.9413687 5 227019 0.197184
## 6 14859.167 0.7560794 5 27071 0.217433
## InterestRate MonthlyLoanPayment TotalDebtToIncomeRatio LoanApproved RiskScore
## 1 0.2275896 419.8060 0.18107720 0 49
## 2 0.2010771 794.0542 0.38985245 0 52
## 3 0.2125480 666.4067 0.46215697 0 52
## 4 0.3009108 1047.5070 0.31309831 0 54
## 5 0.1759902 330.1791 0.07020985 1 36
## 6 0.2176012 385.5771 0.07521129 1 44
## AppID
## 1 20180101
## 2 20180102
## 3 20180103
## 4 20180104
## 5 20180105
## 6 20180106#Mirrors real-world credit scoring systems like FICO, where:
#300–579: Poor
#580–669: Fair
#670–739: Good
#740–799: Very Good
#800–850: Excellent
data <- data %>%
mutate(CreditScoreGroup = case_when(
CreditScore >= 300 & CreditScore <= 579 ~ "Poor",
CreditScore >= 580 & CreditScore <= 669 ~ "Fair",
CreditScore >= 670 & CreditScore <= 739 ~ "Good",
CreditScore >= 740 & CreditScore <= 799 ~ "Very Good",
CreditScore >= 800 & CreditScore <= 850 ~ "Excellent",
TRUE ~ NA_character_ #for other cases
))
head(data)## Age AnnualIncome CreditScore EmploymentStatus EducationLevel Experience
## 1 45 39948 617 Employed Master 22
## 2 38 39709 628 Employed Associate 15
## 3 47 40724 570 Employed Bachelor 26
## 4 58 69084 545 Employed High School 34
## 5 37 103264 594 Employed Associate 17
## 6 37 178310 626 Self-Employed Master 16
## LoanAmount LoanDuration MaritalStatus NumberOfDependents HomeOwnershipStatus
## 1 13152 48 Married 2 Own
## 2 26045 48 Single 1 Mortgage
## 3 17627 36 Married 2 Rent
## 4 37898 96 Single 1 Mortgage
## 5 9184 36 Married 1 Mortgage
## 6 15433 72 Married 0 Rent
## MonthlyDebtPayments CreditCardUtilizationRate NumberOfOpenCreditLines
## 1 183 0.35441792 1
## 2 496 0.08782697 5
## 3 902 0.13741410 2
## 4 755 0.26758714 2
## 5 274 0.32053532 0
## 6 732 0.10221134 5
## NumberOfCreditInquiries DebtToIncomeRatio BankruptcyHistory
## 1 2 0.35833560 0
## 2 3 0.33027367 0
## 3 0 0.24472911 0
## 4 1 0.43624427 0
## 5 0 0.07888421 0
## 6 1 0.25936640 0
## LoanPurpose PreviousLoanDefaults PaymentHistory LengthOfCreditHistory
## 1 Home 0 29 9
## 2 Debt Consolidation 0 21 9
## 3 Education 0 20 22
## 4 Home 0 27 10
## 5 Debt Consolidation 0 26 27
## 6 Debt Consolidation 1 16 19
## SavingsAccountBalance CheckingAccountBalance TotalAssets TotalLiabilities
## 1 7632 1202 146111 19183
## 2 4627 3460 53204 9595
## 3 886 895 25176 128874
## 4 1675 1217 104822 5370
## 5 1555 4981 244305 17286
## 6 2118 1223 67914 40843
## MonthlyIncome UtilityBillsPaymentHistory JobTenure NetWorth BaseInterestRate
## 1 3329.000 0.7249720 11 126928 0.199652
## 2 3309.083 0.9351321 3 43609 0.207045
## 3 3393.667 0.8722406 6 5205 0.217627
## 4 5757.000 0.8961547 5 99452 0.300398
## 5 8605.333 0.9413687 5 227019 0.197184
## 6 14859.167 0.7560794 5 27071 0.217433
## InterestRate MonthlyLoanPayment TotalDebtToIncomeRatio LoanApproved RiskScore
## 1 0.2275896 419.8060 0.18107720 0 49
## 2 0.2010771 794.0542 0.38985245 0 52
## 3 0.2125480 666.4067 0.46215697 0 52
## 4 0.3009108 1047.5070 0.31309831 0 54
## 5 0.1759902 330.1791 0.07020985 1 36
## 6 0.2176012 385.5771 0.07521129 1 44
## AppID CreditScoreGroup
## 1 20180101 Fair
## 2 20180102 Fair
## 3 20180103 Poor
## 4 20180104 Poor
## 5 20180105 Fair
## 6 20180106 Fairunique_values <- unique(data$CreditScoreGroup)
unique_values## [1] "Fair" "Poor" "Good"#Rearrange columns
data <- data %>%
select(AppID, Age, AnnualIncome, CreditScore, CreditScoreGroup, everything())
head(data)## AppID Age AnnualIncome CreditScore CreditScoreGroup EmploymentStatus
## 1 20180101 45 39948 617 Fair Employed
## 2 20180102 38 39709 628 Fair Employed
## 3 20180103 47 40724 570 Poor Employed
## 4 20180104 58 69084 545 Poor Employed
## 5 20180105 37 103264 594 Fair Employed
## 6 20180106 37 178310 626 Fair Self-Employed
## EducationLevel Experience LoanAmount LoanDuration MaritalStatus
## 1 Master 22 13152 48 Married
## 2 Associate 15 26045 48 Single
## 3 Bachelor 26 17627 36 Married
## 4 High School 34 37898 96 Single
## 5 Associate 17 9184 36 Married
## 6 Master 16 15433 72 Married
## NumberOfDependents HomeOwnershipStatus MonthlyDebtPayments
## 1 2 Own 183
## 2 1 Mortgage 496
## 3 2 Rent 902
## 4 1 Mortgage 755
## 5 1 Mortgage 274
## 6 0 Rent 732
## CreditCardUtilizationRate NumberOfOpenCreditLines NumberOfCreditInquiries
## 1 0.35441792 1 2
## 2 0.08782697 5 3
## 3 0.13741410 2 0
## 4 0.26758714 2 1
## 5 0.32053532 0 0
## 6 0.10221134 5 1
## DebtToIncomeRatio BankruptcyHistory LoanPurpose PreviousLoanDefaults
## 1 0.35833560 0 Home 0
## 2 0.33027367 0 Debt Consolidation 0
## 3 0.24472911 0 Education 0
## 4 0.43624427 0 Home 0
## 5 0.07888421 0 Debt Consolidation 0
## 6 0.25936640 0 Debt Consolidation 1
## PaymentHistory LengthOfCreditHistory SavingsAccountBalance
## 1 29 9 7632
## 2 21 9 4627
## 3 20 22 886
## 4 27 10 1675
## 5 26 27 1555
## 6 16 19 2118
## CheckingAccountBalance TotalAssets TotalLiabilities MonthlyIncome
## 1 1202 146111 19183 3329.000
## 2 3460 53204 9595 3309.083
## 3 895 25176 128874 3393.667
## 4 1217 104822 5370 5757.000
## 5 4981 244305 17286 8605.333
## 6 1223 67914 40843 14859.167
## UtilityBillsPaymentHistory JobTenure NetWorth BaseInterestRate InterestRate
## 1 0.7249720 11 126928 0.199652 0.2275896
## 2 0.9351321 3 43609 0.207045 0.2010771
## 3 0.8722406 6 5205 0.217627 0.2125480
## 4 0.8961547 5 99452 0.300398 0.3009108
## 5 0.9413687 5 227019 0.197184 0.1759902
## 6 0.7560794 5 27071 0.217433 0.2176012
## MonthlyLoanPayment TotalDebtToIncomeRatio LoanApproved RiskScore
## 1 419.8060 0.18107720 0 49
## 2 794.0542 0.38985245 0 52
## 3 666.4067 0.46215697 0 52
## 4 1047.5070 0.31309831 0 54
## 5 330.1791 0.07020985 1 36
## 6 385.5771 0.07521129 1 44Exploratory Data Analysis (EDA)
#RiskScore to group if we want to do a classification model
summary(data$RiskScore)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 28.80 46.00 52.00 50.77 56.00 84.00hist(data$RiskScore, breaks = 10, main = "RiskScore Distribution")
# Distribution of RiskScore
ggplot(data, aes(x = RiskScore)) +
geom_histogram(binwidth = 2, color = "black", alpha = 0.9) +
labs(title = "Histogram of RiskScore",
x = "Risk Score",
y = "Count") +
theme_minimal()
# Select numeric columns only
numeric_cols <- data %>% select(where(is.numeric))
# Compute correlation matrix (use complete cases to avoid NAs)
cor_matrix <- cor(numeric_cols, use = "complete.obs")
# Visualize correlation heatmap
corrplot(cor_matrix, method = "color", type = "lower",
tl.cex = 0.5, tl.col = "black",
addCoef.col = "black", number.cex = 0.4,
title = "Correlation Heatmap with LoanApproved",
mar = c(0,0,1,0))
Bivariate Categorical Analysis:
Loan approval vs. credit score group
ggplot(data, aes(x = CreditScoreGroup, fill = factor(LoanApproved))) +
geom_bar(position = "fill") +
labs(title = "Loan Approval by Credit Score Group",
y = "Proportion", x = "Credit Score Group") +
theme_minimal()
Relationship Between Income, Risk, and Loan Approval:
AnnualIncome vs LoanApproved
ggplot(data, aes(x = factor(LoanApproved), y = AnnualIncome)) +
geom_boxplot(fill = "lightblue") +
labs(title = "Annual Income by Loan Approval", x = "Loan Approved", y = "Annual Income")
Relationship Between Income, Risk, and Loan Approval:
RiskScore vs. CreditScore, colored by LoanApproved
ggplot(data, aes(x = CreditScore, y = RiskScore, color = factor(LoanApproved))) +
geom_point(alpha = 0.6) +
labs(title = "Credit Score vs. Risk Score by Loan Approval", x = "Credit Score", y = "Risk Score") +
theme_minimal()
Explore Potential Predictive Features
library(caret)## Loading required package: lattice##
## Attaching package: 'caret'## The following object is masked from 'package:purrr':
##
## liftfeaturePlot(x = data[, c("Age", "AnnualIncome", "CreditScore", "RiskScore")],
y = as.factor(data$LoanApproved),
plot = "density",
auto.key = list(columns = 2))
Data Modeling (Machine learning)
To evaluate the model’s performance, we split the datasets into two parts:
- 80% for Training - used to fit the logistic regression model.
- 20% for Testing - used to evaluate the model on unseen data.
This 80/20 split is a common and widely accepted practice in machine learning to ensure the model generalizes well and does not overfit.
For Binary Classification - We use LoanApproved as target
For Regression - We use RiskScore as target
#Convert LoanApproved to factor. (To treat the numeric values as a binary categorical outcome)
data$LoanApproved <- factor(data$LoanApproved, labels = c("Loan Not Approved", "Loan Approved"))
#Original dataset
table(data$LoanApproved)##
## Loan Not Approved Loan Approved
## 15220 4780#Split the dataset to Training and Testing
install.packages("caTools")## Installing package into 'C:/Users/puter/AppData/Local/R/win-library/4.4'
## (as 'lib' is unspecified)## package 'caTools' successfully unpacked and MD5 sums checked
##
## The downloaded binary packages are in
## C:\Users\puter\AppData\Local\Temp\RtmpEr2lRG\downloaded_packageslibrary(caTools)Binary Classification
set.seed(123) #for reproducibility- the analysis will produce the same results when running it multiple times
split <- sample.split(data$LoanApproved, SplitRatio = 0.8)
train_data <- subset(data, split == TRUE)
test_data <- subset(data, split == FALSE)
kable(summary(train_data$LoanApproved))| x | |
|---|---|
| Loan Not Approved | 12176 |
| Loan Approved | 3824 |
kable(summary(test_data$LoanApproved))| x | |
|---|---|
| Loan Not Approved | 3044 |
| Loan Approved | 956 |
# Standardized the numeric features on training set to improve model performance & comparability for Binary Classification Models
features_to_scale <- c( "LoanAmount", "TotalAssets", "MonthlyIncome", "InterestRate", "TotalDebtToIncomeRatio")
train_data_scaled <- train_data
train_data_scaled[features_to_scale] <- scale(train_data[features_to_scale])
test_data_scaled <- test_data
test_data_scaled[features_to_scale] <- scale(test_data[features_to_scale])Model 1: Logistic Regression
We are using Logistic Regression to model the probability that a loan is approved because the target variable, LoanApproved, is binary and it has only two possible outcomes:
- 1 = Approved
- 0 = Not Approved
Logistic Regression is a classification algorithm that models the relationship between a set of input variables (features) and a binary outcome. It predicts the probability of the loan being approved given the value of the input variables.
Target variable:
- LoanApproved
Features used in the model:
- LoanAmount
- TotalAssets
- MonthlyIncome
- InterestRate
- TotalDebtToIncomeRatio
These features were selected based on their correlation with the target variable, as shown in the heatmap analysis.
# model 1 for binary classification
model_class_1 <- glm(LoanApproved ~ LoanAmount + TotalAssets + MonthlyIncome +
InterestRate + TotalDebtToIncomeRatio,
data = train_data_scaled, family = binomial)## Warning: glm.fit: fitted probabilities numerically 0 or 1 occurredsummary(model_class_1)##
## Call:
## glm(formula = LoanApproved ~ LoanAmount + TotalAssets + MonthlyIncome +
## InterestRate + TotalDebtToIncomeRatio, family = binomial,
## data = train_data_scaled)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -4.42235 0.10676 -41.42 <2e-16 ***
## LoanAmount -1.11394 0.06856 -16.25 <2e-16 ***
## TotalAssets 1.31002 0.04042 32.41 <2e-16 ***
## MonthlyIncome 2.50456 0.08074 31.02 <2e-16 ***
## InterestRate -1.87860 0.05237 -35.87 <2e-16 ***
## TotalDebtToIncomeRatio -3.52042 0.17660 -19.93 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 17597.6 on 15999 degrees of freedom
## Residual deviance: 5844.3 on 15994 degrees of freedom
## AIC: 5856.3
##
## Number of Fisher Scoring iterations: 8# Predict probabilities
pred_probs <- predict(model_class_1, newdata = test_data_scaled, type = "response")
# Predict classes based on a threshold (e.g., 0.5)
pred_classes <- ifelse(pred_probs > 0.5,"Loan Approved","Loan Not Approved" )
pred_classes <- factor(pred_classes, levels = levels(test_data_scaled$LoanApproved))
# Create a confusion matrix
confusion_matrx_1 <- confusionMatrix(factor(pred_classes), factor(test_data_scaled$LoanApproved))
confusion_matrx_1## Confusion Matrix and Statistics
##
## Reference
## Prediction Loan Not Approved Loan Approved
## Loan Not Approved 2912 178
## Loan Approved 132 778
##
## Accuracy : 0.9225
## 95% CI : (0.9138, 0.9306)
## No Information Rate : 0.761
## P-Value [Acc > NIR] : < 2e-16
##
## Kappa : 0.7834
##
## Mcnemar's Test P-Value : 0.01059
##
## Sensitivity : 0.9566
## Specificity : 0.8138
## Pos Pred Value : 0.9424
## Neg Pred Value : 0.8549
## Prevalence : 0.7610
## Detection Rate : 0.7280
## Detection Prevalence : 0.7725
## Balanced Accuracy : 0.8852
##
## 'Positive' Class : Loan Not Approved
## install.packages("pROC")## Installing package into 'C:/Users/puter/AppData/Local/R/win-library/4.4'
## (as 'lib' is unspecified)## package 'pROC' successfully unpacked and MD5 sums checked## Warning: cannot remove prior installation of package 'pROC'## Warning in file.copy(savedcopy, lib, recursive = TRUE): problem copying
## C:\Users\puter\AppData\Local\R\win-library\4.4\00LOCK\pROC\libs\x64\pROC.dll to
## C:\Users\puter\AppData\Local\R\win-library\4.4\pROC\libs\x64\pROC.dll:
## Permission denied## Warning: restored 'pROC'##
## The downloaded binary packages are in
## C:\Users\puter\AppData\Local\Temp\RtmpEr2lRG\downloaded_packageslibrary(pROC)## Type 'citation("pROC")' for a citation.##
## Attaching package: 'pROC'## The following objects are masked from 'package:stats':
##
## cov, smooth, var# Predict probabilities on the test set
pred_probs <- predict(model_class_1, newdata = test_data_scaled, type = "response")
# Generate ROC curve
roc_obj <- roc(test_data_scaled$LoanApproved, pred_probs, levels = c("Loan Approved","Loan Not Approved"))## Setting direction: controls > cases# Plot ROC
plot(roc_obj, col = "blue", main = "ROC Curve")
# AUC
auc_value <- auc(roc_obj)
print(paste("AUC:", round(auc_value, 3)))## [1] "AUC: 0.974"Model 2: Decision Tree
We are using Decision Tree to model the probability that a loan is approved because the target variable, LoanApproved, is binary and it has only two possible outcomes:
- 1 = Approved
- 0 = Not Approved
Decision Tree is a classification algorithm that models the relationship between a set of input variables (features) and a categorical outcome, such as loan approval status. It splits the data into branches based on feature values, creating a tree-like structure that leads to a decision. In binary classification (e.g., “Loan Approved” vs. “Loan Not Approved”), it learns patterns in the features to predict the class label, not just a probability.
Target variable:
- LoanApproved
Features used in the model:
- LoanAmount
- TotalAssets
- MonthlyIncome
- InterestRate
- TotalDebtToIncomeRatio
These features were selected based on their correlation with the target variable, as shown in the heatmap analysis.
# Install library for decision tree algorithm
install.packages("rpart")## Installing package into 'C:/Users/puter/AppData/Local/R/win-library/4.4'
## (as 'lib' is unspecified)## package 'rpart' successfully unpacked and MD5 sums checked## Warning: cannot remove prior installation of package 'rpart'## Warning in file.copy(savedcopy, lib, recursive = TRUE): problem copying
## C:\Users\puter\AppData\Local\R\win-library\4.4\00LOCK\rpart\libs\x64\rpart.dll
## to C:\Users\puter\AppData\Local\R\win-library\4.4\rpart\libs\x64\rpart.dll:
## Permission denied## Warning: restored 'rpart'##
## The downloaded binary packages are in
## C:\Users\puter\AppData\Local\Temp\RtmpEr2lRG\downloaded_packagesinstall.packages("rpart.plot") # Optional: for visualization## Installing package into 'C:/Users/puter/AppData/Local/R/win-library/4.4'
## (as 'lib' is unspecified)## package 'rpart.plot' successfully unpacked and MD5 sums checked
##
## The downloaded binary packages are in
## C:\Users\puter\AppData\Local\Temp\RtmpEr2lRG\downloaded_packageslibrary(rpart)
library(rpart.plot)# model 2 for binary classification
model_class_2 <- rpart(LoanApproved ~ LoanAmount + TotalAssets + MonthlyIncome +
InterestRate + TotalDebtToIncomeRatio,
data = train_data_scaled, method = "class")summary(model_class_2)## Call:
## rpart(formula = LoanApproved ~ LoanAmount + TotalAssets + MonthlyIncome +
## InterestRate + TotalDebtToIncomeRatio, data = train_data_scaled,
## method = "class")
## n= 16000
##
## CP nsplit rel error xerror xstd
## 1 0.37709205 0 1.0000000 1.0000000 0.01410696
## 2 0.04589435 1 0.6229079 0.6555962 0.01202413
## 3 0.01908996 3 0.5311192 0.5462866 0.01114475
## 4 0.01725941 4 0.5120293 0.5431485 0.01111749
## 5 0.01542887 7 0.4602510 0.5109833 0.01083081
## 6 0.01000000 8 0.4448222 0.4864017 0.01060238
##
## Variable importance
## TotalDebtToIncomeRatio MonthlyIncome InterestRate
## 57 28 10
## LoanAmount TotalAssets
## 3 2
##
## Node number 1: 16000 observations, complexity param=0.3770921
## predicted class=Loan Not Approved expected loss=0.239 P(node) =1
## class counts: 12176 3824
## probabilities: 0.761 0.239
## left son=2 (12178 obs) right son=3 (3822 obs)
## Primary splits:
## TotalDebtToIncomeRatio < -0.6732645 to the right, improve=2030.5030, (0 missing)
## MonthlyIncome < 0.4152906 to the left, improve=1788.9430, (0 missing)
## InterestRate < -0.2290712 to the right, improve= 422.4311, (0 missing)
## LoanAmount < -0.3905908 to the right, improve= 274.7782, (0 missing)
## TotalAssets < 2.107242 to the left, improve= 156.7886, (0 missing)
## Surrogate splits:
## MonthlyIncome < 0.5842954 to the left, agree=0.859, adj=0.412, (0 split)
## LoanAmount < -1.350056 to the right, agree=0.763, adj=0.006, (0 split)
##
## Node number 2: 12178 observations, complexity param=0.01725941
## predicted class=Loan Not Approved expected loss=0.09788143 P(node) =0.761125
## class counts: 10986 1192
## probabilities: 0.902 0.098
## left son=4 (8546 obs) right son=5 (3632 obs)
## Primary splits:
## TotalDebtToIncomeRatio < -0.354714 to the right, improve=251.81920, (0 missing)
## MonthlyIncome < 0.0375905 to the left, improve=211.73390, (0 missing)
## InterestRate < -0.8416135 to the right, improve=149.75920, (0 missing)
## TotalAssets < 2.129869 to the left, improve=146.61160, (0 missing)
## LoanAmount < 0.4568591 to the right, improve= 31.43389, (0 missing)
## Surrogate splits:
## MonthlyIncome < -0.07292332 to the left, agree=0.770, adj=0.230, (0 split)
## LoanAmount < -1.164882 to the right, agree=0.705, adj=0.011, (0 split)
## TotalAssets < -0.7779875 to the right, agree=0.702, adj=0.000, (0 split)
##
## Node number 3: 3822 observations, complexity param=0.04589435
## predicted class=Loan Approved expected loss=0.3113553 P(node) =0.238875
## class counts: 1190 2632
## probabilities: 0.311 0.689
## left son=6 (1846 obs) right son=7 (1976 obs)
## Primary splits:
## InterestRate < -0.2238504 to the right, improve=210.898100, (0 missing)
## MonthlyIncome < 0.5706794 to the left, improve=203.141600, (0 missing)
## TotalDebtToIncomeRatio < -0.8485246 to the right, improve=177.069000, (0 missing)
## TotalAssets < 0.2208439 to the left, improve= 47.521030, (0 missing)
## LoanAmount < -0.6595621 to the right, improve= 5.054868, (0 missing)
## Surrogate splits:
## LoanAmount < -0.4034309 to the right, agree=0.574, adj=0.118, (0 split)
## TotalAssets < 1.816473 to the right, agree=0.523, adj=0.012, (0 split)
## TotalDebtToIncomeRatio < -0.8171796 to the right, agree=0.521, adj=0.008, (0 split)
## MonthlyIncome < 3.770106 to the right, agree=0.519, adj=0.004, (0 split)
##
## Node number 4: 8546 observations
## predicted class=Loan Not Approved expected loss=0.03159373 P(node) =0.534125
## class counts: 8276 270
## probabilities: 0.968 0.032
##
## Node number 5: 3632 observations, complexity param=0.01725941
## predicted class=Loan Not Approved expected loss=0.2538546 P(node) =0.227
## class counts: 2710 922
## probabilities: 0.746 0.254
## left son=10 (2475 obs) right son=11 (1157 obs)
## Primary splits:
## InterestRate < -0.5882456 to the right, improve=203.577800, (0 missing)
## TotalAssets < 1.047169 to the left, improve=101.485800, (0 missing)
## MonthlyIncome < 0.1459273 to the left, improve= 78.070170, (0 missing)
## TotalDebtToIncomeRatio < -0.5248973 to the right, improve= 37.063880, (0 missing)
## LoanAmount < 0.4573817 to the right, improve= 4.511907, (0 missing)
## Surrogate splits:
## LoanAmount < -1.339717 to the right, agree=0.683, adj=0.006, (0 split)
## TotalDebtToIncomeRatio < -0.3553237 to the left, agree=0.683, adj=0.004, (0 split)
## MonthlyIncome < 2.418231 to the left, agree=0.682, adj=0.003, (0 split)
## TotalAssets < 6.512311 to the left, agree=0.682, adj=0.002, (0 split)
##
## Node number 6: 1846 observations, complexity param=0.04589435
## predicted class=Loan Approved expected loss=0.4832069 P(node) =0.115375
## class counts: 892 954
## probabilities: 0.483 0.517
## left son=12 (1131 obs) right son=13 (715 obs)
## Primary splits:
## MonthlyIncome < 1.174252 to the left, improve=172.703400, (0 missing)
## TotalDebtToIncomeRatio < -0.8551 to the right, improve=156.231700, (0 missing)
## InterestRate < 0.7787038 to the right, improve= 54.585080, (0 missing)
## TotalAssets < 0.2214176 to the left, improve= 49.830070, (0 missing)
## LoanAmount < -0.1487182 to the left, improve= 5.963172, (0 missing)
## Surrogate splits:
## TotalDebtToIncomeRatio < -0.8727742 to the right, agree=0.736, adj=0.317, (0 split)
## LoanAmount < 0.0217124 to the left, agree=0.719, adj=0.274, (0 split)
## InterestRate < 3.048491 to the left, agree=0.615, adj=0.007, (0 split)
##
## Node number 7: 1976 observations
## predicted class=Loan Approved expected loss=0.1508097 P(node) =0.1235
## class counts: 298 1678
## probabilities: 0.151 0.849
##
## Node number 10: 2475 observations
## predicted class=Loan Not Approved expected loss=0.1393939 P(node) =0.1546875
## class counts: 2130 345
## probabilities: 0.861 0.139
##
## Node number 11: 1157 observations, complexity param=0.01725941
## predicted class=Loan Not Approved expected loss=0.4987035 P(node) =0.0723125
## class counts: 580 577
## probabilities: 0.501 0.499
## left son=22 (759 obs) right son=23 (398 obs)
## Primary splits:
## MonthlyIncome < 0.2026565 to the left, improve=75.861990, (0 missing)
## TotalAssets < 0.3525364 to the left, improve=44.057170, (0 missing)
## InterestRate < -1.307246 to the right, improve=38.160860, (0 missing)
## TotalDebtToIncomeRatio < -0.58997 to the right, improve=21.561770, (0 missing)
## LoanAmount < -0.01348603 to the left, improve= 5.874888, (0 missing)
## Surrogate splits:
## LoanAmount < -0.06171124 to the left, agree=0.774, adj=0.342, (0 split)
## TotalAssets < -0.7451932 to the right, agree=0.658, adj=0.005, (0 split)
## InterestRate < -1.937091 to the right, agree=0.658, adj=0.005, (0 split)
## TotalDebtToIncomeRatio < -0.643018 to the right, agree=0.657, adj=0.003, (0 split)
##
## Node number 12: 1131 observations, complexity param=0.01908996
## predicted class=Loan Not Approved expected loss=0.3448276 P(node) =0.0706875
## class counts: 741 390
## probabilities: 0.655 0.345
## left son=24 (1020 obs) right son=25 (111 obs)
## Primary splits:
## TotalAssets < 1.058802 to the left, improve=57.664480, (0 missing)
## MonthlyIncome < 0.572567 to the left, improve=38.278080, (0 missing)
## InterestRate < 0.7787038 to the right, improve=32.074390, (0 missing)
## TotalDebtToIncomeRatio < -0.8485246 to the right, improve=29.997210, (0 missing)
## LoanAmount < -0.4383308 to the right, improve= 8.398313, (0 missing)
## Surrogate splits:
## MonthlyIncome < -0.7813846 to the right, agree=0.904, adj=0.018, (0 split)
##
## Node number 13: 715 observations
## predicted class=Loan Approved expected loss=0.2111888 P(node) =0.0446875
## class counts: 151 564
## probabilities: 0.211 0.789
##
## Node number 22: 759 observations, complexity param=0.01542887
## predicted class=Loan Not Approved expected loss=0.3675889 P(node) =0.0474375
## class counts: 480 279
## probabilities: 0.632 0.368
## left son=44 (616 obs) right son=45 (143 obs)
## Primary splits:
## TotalAssets < 0.3154161 to the left, improve=40.426830, (0 missing)
## InterestRate < -1.199029 to the right, improve=32.862340, (0 missing)
## MonthlyIncome < -0.2135006 to the left, improve=12.154800, (0 missing)
## TotalDebtToIncomeRatio < -0.4056125 to the right, improve=10.312910, (0 missing)
## LoanAmount < -0.6725888 to the right, improve= 3.949578, (0 missing)
##
## Node number 23: 398 observations
## predicted class=Loan Approved expected loss=0.2512563 P(node) =0.024875
## class counts: 100 298
## probabilities: 0.251 0.749
##
## Node number 24: 1020 observations
## predicted class=Loan Not Approved expected loss=0.2921569 P(node) =0.06375
## class counts: 722 298
## probabilities: 0.708 0.292
##
## Node number 25: 111 observations
## predicted class=Loan Approved expected loss=0.1711712 P(node) =0.0069375
## class counts: 19 92
## probabilities: 0.171 0.829
##
## Node number 44: 616 observations
## predicted class=Loan Not Approved expected loss=0.288961 P(node) =0.0385
## class counts: 438 178
## probabilities: 0.711 0.289
##
## Node number 45: 143 observations
## predicted class=Loan Approved expected loss=0.2937063 P(node) =0.0089375
## class counts: 42 101
## probabilities: 0.294 0.706# Predict classes directly from the decision tree
pred_classes <- predict(model_class_2, newdata = test_data_scaled, type = "class")
# Create confusion matrix for Decision Tree Model
confusion_matrx_2 <- confusionMatrix(pred_classes, test_data_scaled$LoanApproved)
confusion_matrx_2## Confusion Matrix and Statistics
##
## Reference
## Prediction Loan Not Approved Loan Approved
## Loan Not Approved 2891 302
## Loan Approved 153 654
##
## Accuracy : 0.8862
## 95% CI : (0.876, 0.8959)
## No Information Rate : 0.761
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.6696
##
## Mcnemar's Test P-Value : 3.967e-12
##
## Sensitivity : 0.9497
## Specificity : 0.6841
## Pos Pred Value : 0.9054
## Neg Pred Value : 0.8104
## Prevalence : 0.7610
## Detection Rate : 0.7228
## Detection Prevalence : 0.7983
## Balanced Accuracy : 0.8169
##
## 'Positive' Class : Loan Not Approved
## Compare Model Performance
# Extract metrics from confusion matrix 1
acc_1 <- confusion_matrx_1$overall["Accuracy"]
prec_1 <- confusion_matrx_1$byClass["Precision"]
rec_1 <- confusion_matrx_1$byClass["Recall"]
f1_1 <- confusion_matrx_1$byClass["F1"]
# Extract metrics from confusion matrix 2
acc_2 <- confusion_matrx_2$overall["Accuracy"]
prec_2 <- confusion_matrx_2$byClass["Precision"]
rec_2 <- confusion_matrx_2$byClass["Recall"]
f1_2 <- confusion_matrx_2$byClass["F1"]
# Combine into a data frame for comparison
model_eval <- data.frame(
Metric = c("Accuracy", "Precision", "Recall", "F1 Score"),
Model_1 = c(acc_1, prec_1, rec_1, f1_1),
Model_2 = c(acc_2, prec_2, rec_2, f1_2)
)
kable(model_eval)| Metric | Model_1 | Model_2 | |
|---|---|---|---|
| Accuracy | Accuracy | 0.9225000 | 0.8862500 |
| Precision | Precision | 0.9423948 | 0.9054181 |
| Recall | Recall | 0.9566360 | 0.9497372 |
| F1 | F1 Score | 0.9494620 | 0.9270483 |
library(ggplot2)
library(reshape2)##
## Attaching package: 'reshape2'## The following object is masked from 'package:tidyr':
##
## smithsmelted_eval <- melt(model_eval, id.vars = "Metric")
ggplot(melted_eval, aes(x = Metric, y = value, fill = variable)) +
geom_bar(stat = "identity", position = "dodge") +
labs(title = "Model Evaluation Comparison", y = "Score", x = "Metric") +
theme_minimal() +
scale_fill_manual(values = c("Model_1" = "steelblue", "Model_2" = "tomato"))
Model Performance Analysis
Accuracy:
- Logistic Regression (92.25%) outperforms Decision Tree (88.63%)
- Correctly classifies more overall instances.
Precision (Positive Predictive Value):
- Logistic Regression: 94.24% of its positive predictions are correct.
- Decision Tree: 90.54% precision.
- Logistic Regression is better at avoiding false positives.
Recall (Sensitivity / True Positive Rate):
- Logistic Regression: 95.66%, slightly higher than Decision Tree: 94.97%.
- Both models are strong at identifying actual positives, but Logistic Regression is marginally better.
F1 Score (Harmonic mean of Precision and Recall):
- Logistic Regression: 94.95%, higher than Decision Tree: 92.70%.
- Indicates Logistic Regression strikes a better balance between precision and recall.
Conclusion
Logistic Regression model performs better overall across all key evaluation metrics—Accuracy, Precision, Recall, and F1 Score. It not only makes more correct predictions overall but also better balances between identifying positive cases and avoiding false positives.
Regression Modeling
set.seed(123) #for reproducibility- the analysis will produce the same results when running it multiple times
reg_split <- sample.split(data$RiskScore, SplitRatio = 0.8)
reg_train_data <- subset(data, split == TRUE)
reg_test_data <- subset(data, split == FALSE)
summary(reg_train_data$RiskScore)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 28.80 46.00 52.00 50.76 56.00 79.00summary(reg_test_data$RiskScore)## Min. 1st Qu. Median Mean 3rd Qu. Max.
## 28.80 46.00 52.00 50.78 56.00 84.00# Standardized the numeric features on training set to improve model performance & comparability for Regression Models
reg_features_to_scale <- c( "CreditScore", "DebtToIncomeRatio", "BankruptcyHistory", "PreviousLoanDefaults", "TotalAssets", "MonthlyIncome", "NetWorth", "InterestRate", "TotalDebtToIncomeRatio")
reg_train_data_scaled <- reg_train_data
reg_train_data_scaled[reg_features_to_scale] <- scale(reg_train_data[reg_features_to_scale])
reg_test_data_scaled <- reg_test_data
reg_test_data_scaled[reg_features_to_scale] <- scale(reg_test_data[reg_features_to_scale])Model 3: Linear Regression
We are using Linear Regression to model and predict the RiskScore, a continuous numeric variable that quantifies the financial risk associated with a loan applicant.
Linear Regression is a fundamental regression technique that models the relationship between one dependent variable (in this case, RiskScore) and one or more independent variables (predictors) by fitting a linear equation to the observed data. It assumes a linear relationship between the input features and the target variable.
The goal is to estimate the coefficients that minimize the difference between the actual and predicted values of RiskScore.
Target variable:
- RiskScore
Features used in the model:
- CreditScore
- DebtToIncomeRatio
- BankruptcyHistory
- PreviousLoanDefaults
- TotalAssets
- MonthlyIncome
- NetWorth
- InterestRate
- TotalDebtToIncomeRatio
These features were selected based on domain relevance and correlation strength with RiskScore, as observed during exploratory data analysis. The model was trained on 80% of the data and evaluated on the remaining 20% using RMSE, MAE, and R-squared metrics to assess its predictive performance.
# Build the linear regression model
model_reg_1 <- lm(RiskScore ~ CreditScore + DebtToIncomeRatio + BankruptcyHistory +
PreviousLoanDefaults + TotalAssets + MonthlyIncome + NetWorth +
InterestRate + TotalDebtToIncomeRatio,
data = reg_train_data_scaled)
# Summary of the model
summary(model_reg_1)##
## Call:
## lm(formula = RiskScore ~ CreditScore + DebtToIncomeRatio + BankruptcyHistory +
## PreviousLoanDefaults + TotalAssets + MonthlyIncome + NetWorth +
## InterestRate + TotalDebtToIncomeRatio, data = reg_train_data_scaled)
##
## Residuals:
## Min 1Q Median 3Q Max
## -14.991 -2.478 0.322 2.735 44.094
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 50.76305 0.03302 1537.518 <2e-16 ***
## CreditScore -0.61732 0.04140 -14.910 <2e-16 ***
## DebtToIncomeRatio 2.49912 0.03303 75.665 <2e-16 ***
## BankruptcyHistory 2.94551 0.03303 89.185 <2e-16 ***
## PreviousLoanDefaults 2.03487 0.03303 61.609 <2e-16 ***
## TotalAssets 0.04880 0.15825 0.308 0.758
## MonthlyIncome -3.31265 0.03931 -84.266 <2e-16 ***
## NetWorth -2.44938 0.15824 -15.478 <2e-16 ***
## InterestRate 1.39543 0.04145 33.666 <2e-16 ***
## TotalDebtToIncomeRatio 0.64448 0.03953 16.305 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 4.176 on 15990 degrees of freedom
## Multiple R-squared: 0.7124, Adjusted R-squared: 0.7122
## F-statistic: 4400 on 9 and 15990 DF, p-value: < 2.2e-16# Predict on the test set
predictions <- predict(model_reg_1, newdata = test_data_scaled)
# View the first few predictions
head(predictions)## 4 7 11 14 22 27
## -243879.18 -53523.05 -315357.43 -203074.60 -13974.82 -41265.75# Load required library
install.packages("Metrics")## Installing package into 'C:/Users/puter/AppData/Local/R/win-library/4.4'
## (as 'lib' is unspecified)## package 'Metrics' successfully unpacked and MD5 sums checked
##
## The downloaded binary packages are in
## C:\Users\puter\AppData\Local\Temp\RtmpEr2lRG\downloaded_packageslibrary(Metrics)##
## Attaching package: 'Metrics'## The following object is masked from 'package:pROC':
##
## auc## The following objects are masked from 'package:caret':
##
## precision, recall# Actual vs Predicted
actual <- reg_test_data_scaled$RiskScore
predicted <- predictions
# Calculate performance metrics
mse <- mse(actual, predicted)
rmse <- rmse(actual, predicted)
mae <- mae(actual, predicted)
r_squared <- 1 - sum((actual - predicted)^2) / sum((actual - mean(actual))^2)
# Print the metrics
cat("MSE:", round(mse, 2), "\n")## MSE: 123504775705cat("RMSE:", round(rmse, 2), "\n")## RMSE: 351432.5cat("MAE:", round(mae, 2), "\n")## MAE: 178214.2cat("R-squared:", round(r_squared, 4), "\n")## R-squared: -2055390643Model 4: Random Forest Regression
We are using Random Forest Regression to model and predict the RiskScore, a continuous numeric variable that quantifies the financial risk associated with a loan applicant.
Random Forest is an ensemble learning algorithm that builds multiple decision trees during training and averages their predictions for regression tasks. This approach helps reduce overfitting and improves generalization performance, especially when the relationship between variables is complex and potentially non-linear.
The model captures interactions and nonlinear patterns in the data by leveraging the “wisdom of the crowd” from multiple trees, making it more robust than a single linear model.
Target variable:
- RiskScore
Features used in the model:
- CreditScore
- DebtToIncomeRatio
- BankruptcyHistory
- PreviousLoanDefaults
- TotalAssets
- MonthlyIncome
- NetWorth
- InterestRate
- TotalDebtToIncomeRatio
The model was trained using 80% of the data and evaluated on the remaining 20%. Evaluation metrics such as RMSE, MAE, and R-squared were used to assess model accuracy. Additionally, feature importance was analyzed to understand the contribution of each variable to the predictions.
install.packages("randomForest")## Installing package into 'C:/Users/puter/AppData/Local/R/win-library/4.4'
## (as 'lib' is unspecified)## package 'randomForest' successfully unpacked and MD5 sums checked
##
## The downloaded binary packages are in
## C:\Users\puter\AppData\Local\Temp\RtmpEr2lRG\downloaded_packageslibrary(randomForest)## randomForest 4.7-1.2## Type rfNews() to see new features/changes/bug fixes.##
## Attaching package: 'randomForest'## The following object is masked from 'package:dplyr':
##
## combine## The following object is masked from 'package:ggplot2':
##
## margin# Build random forest regressor
model_rf <- randomForest(RiskScore ~ CreditScore + DebtToIncomeRatio + BankruptcyHistory +
PreviousLoanDefaults + TotalAssets + MonthlyIncome + NetWorth +
InterestRate + TotalDebtToIncomeRatio,
data = reg_train_data_scaled, ntree = 100)# Predict
predictions_rf <- predict(model_rf, newdata = reg_test_data_scaled)# Evaluate
mse_rf <- mse(reg_test_data_scaled$RiskScore, predictions_rf)
rmse_rf <- rmse(reg_test_data_scaled$RiskScore, predictions_rf)
mae_rf <- mae(reg_test_data_scaled$RiskScore, predictions_rf)
r2_rf <- 1 - sum((reg_test_data_scaled$RiskScore - predictions_rf)^2) /
sum((reg_test_data_scaled$RiskScore - mean(reg_test_data_scaled$RiskScore))^2)cat("Random Forest Regression Performance:\n")## Random Forest Regression Performance:cat("MSE:", round(mse_rf, 2), "\n")## MSE: 10.94cat("RMSE:", round(rmse_rf, 2), "\n")## RMSE: 3.31cat("MAE:", round(mae_rf, 2), "\n")## MAE: 2.4cat("R-squared:", round(r2_rf, 4), "\n")## R-squared: 0.8179Model Performance Analysis
# Combine into one data frame for comparison
actual <- reg_test_data_scaled$RiskScore
pred_lm <- predict(model_reg_1, newdata = reg_test_data_scaled)
pred_rf <- predict(model_rf, newdata = reg_test_data_scaled)
viz_data <- data.frame(
Actual = rep(actual, 2),
Predicted = c(pred_lm, pred_rf),
Model = rep(c("Linear Regression", "Random Forest"), each = length(actual))
)library(ggplot2)
ggplot(viz_data, aes(x = Actual, y = Predicted)) +
geom_point(alpha = 0.5) +
geom_abline(color = "blue", linetype = "dashed") +
facet_wrap(~ Model) +
labs(title = "Actual vs. Predicted RiskScore",
x = "Actual", y = "Predicted") +
theme_minimal()
Linear Regression (Left plot): * The points are fairly close to the diagonal, but there’s more vertical spread (error). * This suggests the model captures the overall trend but has larger residuals, especially for extreme values (under/over-predicts a bit).
Random Forest (Right plot): * The points are tighter and more aligned along the dashed line. * This means the predictions are more accurate, especially across a wide range of RiskScore values.
The reduced spread indicates less variance and better generalization.
Conclusion: Random Forest performs better than Linear Regression in predicting RiskScore.
library(Metrics)
results <- data.frame(
Model = c("Linear Regression", "Random Forest"),
RMSE = c(rmse(actual, pred_lm), rmse(actual, pred_rf)),
MAE = c(mae(actual, pred_lm), mae(actual, pred_rf)),
R2 = c(
1 - sum((actual - pred_lm)^2) / sum((actual - mean(actual))^2),
1 - sum((actual - pred_rf)^2) / sum((actual - mean(actual))^2)
)
)# RMSE bar chart
ggplot(results, aes(x = Model, y = RMSE, fill = Model)) +
geom_col() +
labs(title = "RMSE Comparison", y = "RMSE") +
theme_minimal()
The Random Forest model achieved a lower RMSE, indicating that it makes more accurate predictions of RiskScore on the test set compared to Linear Regression.
# R-squared bar chart
ggplot(results, aes(x = Model, y = R2, fill = Model)) +
geom_col() +
labs(title = "R-squared Comparison", y = "R-squared") +
theme_minimal()
The Random Forest model explains a greater proportion of the variance in RiskScore, suggesting it has stronger predictive power and better fits the data than the Linear Regression model.
Feature Importance (Random Forest)
importance <- importance(model_rf)
importance_df <- data.frame(Feature = rownames(importance), Importance = importance[, 1])
# Plot
ggplot(importance_df, aes(x = reorder(Feature, Importance), y = Importance)) +
geom_col(fill = "steelblue") +
coord_flip() +
labs(title = "Feature Importance (Random Forest)",
x = "Features", y = "Importance") +
theme_minimal()
Key Insights:
- TotalDebtToIncomeRatio is the most important predictor, indicating that applicants with higher debt burdens relative to income are more likely to have elevated risk scores.
- MonthlyIncome also plays a crucial role — it suggests that income level is strongly predictive of an applicant’s financial stability.
- BankruptcyHistory and DebtToIncomeRatio follow closely, which aligns well with real-world expectations: historical financial behavior and current debt load significantly influence credit risk.
- Other features like NetWorth, InterestRate, and CreditScore also contribute, but with lesser influence.
Discussion:
Based on both visual and quantitative evaluation metrics, the Random Forest Regression model outperforms the Linear Regression model. It shows a lower RMSE and a higher R², indicating improved accuracy and better generalization. The scatter plots of predicted vs. actual values also support this conclusion, as the Random Forest predictions align more closely along the ideal diagonal line.
The Random Forest model identified TotalDebtToIncomeRatio and MonthlyIncome as the most influential features in predicting RiskScore, followed by BankruptcyHistory and DebtToIncomeRatio. This aligns with financial logic — higher debt obligations and lower income levels generally indicate higher financial risk. These insights can inform future loan approval strategies and highlight which applicant attributes require closer scrutiny.
Conclusion
This project successfully demonstrated the power and versatility of the R ecosystem for executing an end to end data science workflow. Essential libraries like tidyverse and dplyr were pivotal for data preparation, while ggplot2 and corrplot enabled crucial data exploration. The modeling phase was effectively handled using packages such as caret, rpart, and randomForest, allowing for the development and rigorous evaluation of multiple machine learning models that successfully met the project’s objectives.
The analysis identified clear winning models for each predictive task and uncovered the key drivers of financial risk. For binary classification, Logistic Regression proved superior with an impressive 92.25% accuracy. In predicting the applicant’s RiskScore, Random Forest Regression was exceptionally effective, explaining 81.8% of the R² and highlighting that an applicant’s TotalDebtToIncomeRatio and MonthlyIncome are the most influential predictive features. These findings provide a practical, data driven framework to enhance the accuracy and fairness of future loan approval decisions.
Summary of R Libraries/Packages
The R ecosystem provides a rich set of tools for every data science task. For our loan approval analysis, we utilized the following key packages to handle data pre-processing, create insightful visualizations, and build our predictive models.
Data Manipulation & Visualization
| Libraries/Packages | Description |
|---|---|
| tidyverse |
A collection of essential R packages for data science, including dplyr and ggplot2.
|
| dplyr |
Used for data manipulation tasks like mutate and
select.
|
| knitr |
Used for creating formatted tables using kable().
|
| reshape2 |
Used for restructuring data, specifically for creating the model comparison plot. |
| skimr | Used for generating summary statistics of the data. |
| caTools |
Used for splitting the data into training and testing sets using
sample.split().
|
| ggplot2 |
The primary tool used for creating plots: histograms, boxplots, bar charts, scatter plots. |
| corrplot | Used specifically to create the correlation heatmap. |
| rpart.plot | Used to visualize the decision tree model. |
Machine Learning & Modeling
| Libraries/Packages | Description |
|---|---|
| caret |
A comprehensive framework for model training, feature selection, and
evaluation. It was used for featurePlot.
|
| rpart | The package used to build the Decision Tree model. |
| randomForest | The package used to build the Random Forest model. |
| pROC | Used to generate and analyze the ROC curve and calculate the AUC for the logistic regression model. |
| Metrics | Used to calculate regression model performance metrics like MSE, RMSE, and MAE. |