Loan Analysis Predicting Defaulters

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 04 | Apr 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1442
Loan Analysis Predicting Defaulters
Mudit Manish Agarwal 1, Harshal Mahendra Shirke 2, Vivek Prafullbhai Vadhiya 3,
Manya Gidwani 4
1,2,3Student, Department of Information Technology, Shah & Anchor Kutchhi Engineering College, Mumbai, India
4 Professor, Department of Information Technology, Shah & Anchor Kutchhi Engineering College, Mumbai, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Due to the advancements in the domain of
Artificial Intelligence and Data Science, its utilization is
becoming more common in every possible domain.
Nowadays, the majority of the industries make use of AI
and its applications in some or the other way. Taking the
advantage of the field of Data Science results in creating
effective and modern applications, products irrespective
of the domain. One of the industrieswheretheapplication
of AI and Data Science is proving to be effective is the
Finance Industry commonly known astheBankingSector.
Banks face severe losses due to the loan defaults made by
the client and hence to overcome thisproblem,thereliesa
need to create a credit risk scoring model which can
analyze and predict the loan defaults. Hence, with the
help of Machine Learning, weaimtocreatea Loan Default
Analysis model which could predict the loan defaults and
integrate the model into a web application fortheuserfor
easy usability.
Key Words: LoanDefault,MachineLearning,datamining,
prediction, web application.
1. INTRODUCTION
Due to the Covid-19 Pandemic, there is a huge loss of
capital caused to the banking sector, financial institutions,
small scale finance companies, etc. Nearly a year and a half,
everything has been shut down thereby leaving people with
no source of income. Due to this reason, there has been a
significant increase in the loan defaults made by the client.
Now, in this new normal, there is a need especially for the
banks to strengthen their loan sanctioning system. Since
banks may face huge losses due to the defaultsmade bytheir
clients which increases the rejection rate of the loan
applicants. This affectsthe bank'soverall reputationand also
at the same time, due to the rejection of new loan applicants
it causes huge financial loss to the bank since, the most
common type of unsecured loans are debt consolidation,
credit-card loans, student loans, and personal loans [1].
The Loan Analysis Predicting Defaulters (LAPD) is an
attempt at creating a better credit risk scoring model which
can correctly identify which applicant will be a defaulter in
the future. This is done by analyzing the historic data and
identifying the patterns.Sucha model wouldminimizecredit
risk and prevent the clients who are capable of repayment
from getting rejected. Various classification algorithmssuch
as Logistic Regression, Decision Tree, Random Forest, etc.
have been applied to build various models and compare
them to find out the best accuracy [4]. The source of the
dataset is Kaggle which is a highly imbalanced dataset. Data
Balancing techniquessuchasSMOTE,SMOTEENN havebeen
implemented to balancethe dataset.Importantfeaturessuch
as loan_amount, term, home_ownership,issue_date_year, etc
have been extracted on which the target attribute depends
for prediction. Once the model creation and comparison are
done, the model which gives better accuracy is integrated
into the website for the end-user
2. BUSINESS PROBLEM
Banks may suffer significant lossesasaresultofcustomer
defaults, which raises the rejection rate of loan applications.
To address this issue, we require a more accurate credit risk
assessment algorithm that can predict which applicants will
default in the future. This will be accomplished by analyzing
previous data and discovering trends. This methodology
would reduce credit riskand prevent clientswhoarecapable
of repayment from being turned down [3].
Fig. 2.1 Purpose of loan
3. DATA DESCRIPTION
The Data Set was taken from Kaggle The dataset includes
entire loan data for all loans granted between 2007 and
2015, including current loan status(Current,Late,FullyPaid,

etc.) and most recent payment information. There are
8,55,969 items in our dataset, with 73 attributes including
the target variable. Furthermore, the dataset is extremely
imbalanced, with 46467 entries of failed loans. This dataset
contains a variety of attributes, including category, numeric,
and date data. The number of defaulters increased
significantly between 2012 and 2014, with the LIBOR
Scandal, Hurricane Sandy, and Hostess Files for Bankruptcy
being some of the primary factors. The major reason for a
borrower's request for a loan is the loan purpose. Debt
consolidation is the most common reason for taking out a
loan, followed by credit card debt.
Some important features from the dataset
• loan_amnt - Amount of money requested by the borrower.
• int_rate - The interest rate on the loan.
• grade - Loan grade with categories A, B, C, D, E, F, G.
• annual_inc - Borrowers annual income.
• purpose - The primary purpose of borrowing.
• installments – Monthly amount payments for the opted
loan.
• term – duration of the loan until it’s paid off
• to – the ratio of your gross monthly income that goes to
paying your monthly debt payments [6].
Fig. 3.1 Understanding the dataset
4. DATA PRE-PROCESSING
The data wereinspected,cleaned,andpreparedasfollows
before data mining model analysis: Changing the names of
the variables — We changed the names of the variables to
suit our needs and domain expertise. Retrieved the values of
variables such as emp lengthand term, which had values in a
string from which we extracted the numeric component.
E.g.: -'10+ years' -> 10;
'2 years' -> 2;
'1 year' -> 1;
'<1 year' -> 0;
We did the same thing for the term with string values.
For example: - 36 months -> 36
60 months -> 60
Values from one variable are replaced with values from
another variable. Based on the application type, the income
value of the borrower is replaced with the income value of
the co-borrower. If the application type is INDIVIDUAL, the
borrower's income will be preserved, but if the application
type is JOINT, the current value of the borrower'sincomewill
be substituted with joint income. while joint income is the
sum of the borrower's and co-income. borrower's Dti/ratio
inc exp borrower and dti joint/ratio inc exp joint follow the
same technique.
4.1 Label Encoders
The Label Encoder is a method for converting category
data to numeric variables. We used the Label encoder
approach to categorize the aforementioned variables, which
are categorical. We have two functions in this technique: fit
and transform.
Label Encoder's Steps:
1. Obtaining the one-of-a-kind values
2. Putting them in ascending order is a good idea.
3. Starting with 0,1,2, and so on, mapping the values.
4. All three processes will be completed by the Fit
function.
5. The data values in the data framewill bereplacedby
the transform function.
4.2 Train Test Split
Splitting the data into train and test in the ratio7:3wherethe
training part contains 70 % of the data and Testing part
contains 30% of the data Train Size – 598978 records with 8
features Test Size – 256991 records with 8 features. The
Dataset was highly Imbalanced with variable 0 (Non –
Defaulters) = 809502 and 1 (Defaulters) = 46467. In order,
the balance this day out we have used two techniques which
are Smote and Smote ENN [3].
4.3 Balancing the dataset
4.3.1 SMOTE
Imbalanced classification entails creating predictionmodels
for datasets with a significant class imbalance. When
working with unbalanced datasets, the difficultyisthatmost
machine learning approaches will overlook the minority
class, resulting in poor performance, even though
performance on the minority class is often the most

significant. Oversampling the minority class is one way to
deal with unbalanced datasets. Duplicating instances in the
minority class is the easiest way, but these examples don't
provide any new information to the model [7]. Instead, new
instances may be created by synthesizing old ones. The
Synthetic Minority Oversampling Technique, or SMOTE for
short, is a kind of data augmentation for the minority
population. Imbalanced classification entails creating
prediction models for datasets with a significant class
imbalance. When working with unbalanced datasets, the
difficulty is that most machine learning approaches will
overlook the minority class, resulting in poor performance,
even though performance on the minority class is often the
most significant. Oversampling the minorityclassisoneway
to deal with unbalanced datasets. Duplicating instances in
the minority class is the easiest way, but these examples
don't provide any new information to the model. Instead,
new instances may be created by synthesizing old ones. The
Synthetic Minority Oversampling Technique, or SMOTE for
short, is a kind of data augmentation for the minority
population.
4.3.2 SMOTE ENN
This method, developed by Batista et al. (2004), combines
the ability of SMOTE to generate synthetic examples for
minority classes with the ability of ENN to delete some
observations from both classes that are identified as having
different classes between the observation's class and its K-
nearest neighbor majority class. The following is a
description of the SMOTE-ENN procedure.
(SMOTE begins) Select data at random from the minority
class. Calculate the distance between the randomly
generated data and the k closest neighbors. Add theresultto
the minority class as a synthetic sample by multiplying the
difference by a random value between 0 and 1. Repeat steps
2–3 until the required minority class proportion isachieved.
(SMOTE comes to an end) (ENN begins) CalculateKtobe the
number of closest neighbors. If K can't be determined,it'll be
3. Find the observation's K-nearest neighbor from the
dataset's other observations, then return the majority class
from the K-nearest neighbor. If the observation's class and
its K-nearest neighbor's majority class are not the same, the
observation and its K-nearest neighbor are removed from
the dataset. Repeat steps 2 and 3 until each class has the
required proportion of students.(TheENN comestoa close.)
[3].
After using the techniques the data was balanced with equal
parts using smote and in the ratio 6:4 by using smote ENN.
Fig. 4.1 Dataset after balancing
4.4. Algorithms
4.4.1 Logistic Regression
The method of modeling the likelihood of a discrete result
given an input variable is known as logistic regression. The
most frequent logistic regression models have a binary
result, which might be true or false, yes or no, and so forth.
Multinomial logistic regression can be used to model
situations with more than two discrete outcomes. Logistic
regression is a handy analytical tool for determining if a
fresh sample fits best into a category in classification tasks.
Because components of cyber security, such as threat
detection, are classification issues, logistic regression is a
valuable analytic tool [2].
4.4.2 Decision Tree
For classification and regression, Decision Trees (DTs) are a
non-parametric supervisedlearningapproach.Theobjective
is to learn basic decision rules from data attributes to
develop a model that predicts the value of a target variable.
A tree is an approximation of a piecewise constant. Decision
Trees are a form of supervised machine learning in which
the data is continually split according to a parameter (you
describe what the input is and what the related output is in
the training data). Two entities, decision nodes, and leaves
can be used to explain the tree. The decisions or ultimate
outcomes are represented by the leaves. And the data is
separated at the decision nodes. At first, we consider the
entire training set to be the root. Categorical feature values
are desired. If the values are continuous, they must be
discretized before the model can be built. Records are
distributed recursively based on attributevalues.Astheroot
of the internal node, we apply statistical approaches to rank
characteristics.

4.4.3 Ada Boost
The AdaBoost algorithm, short for Adaptive Boosting, is a
Boosting approach usedinMachineLearningasanEnsemble
Method. The weights are re-allocated to each instance, with
larger weights applied to improperly identified instances.
This is termed Adaptive Boosting. In supervised learning,
boost is used to decrease bias and variation. It is based on
the notion of successive learning. Each succeeding student,
except the first, is produced from previouslygrownlearners.
In other words, weak students are transformed into strong
students [6]. With a little modification,theAdaBoostmethod
operates on the same idea as boosting. Adaptive Boosting is
an effective ensemble approach for both classification and
regression issues. It is most commonly used to solve
categorization difficulties. It outperformsall othermodelsin
terms of model correctness, which can be verified by
following the steps in order. To apply the boost and
implement AdaBoost, one can start with decision trees and
then move on to random forests. As we progressthroughthe
steps outlined above, accuracy improves. The weight-
assigning approach used after each iteration distinguishes
the AdaBoost algorithm from all other boosting algorithms,
which is its strongest feature.
4.4.4 Random Forest
Random forest is a supervised machine learning algorithm
that is commonly used to solve classification and regression
issues. It creates decision trees from several samples, using
the majority vote for classification and the average for
regression. One of the most essential characteristics of the
Random Forest Algorithm is that it can handle data sets with
both continuous and categorical variables, as in regression
and classification. For classification difficulties, it produces
superior results. Random Forest Algorithm Characteristics-
1) It outperforms the decision tree algorithm in terms of
accuracy.
2) It is a useful tool for dealing with missing data.
3) Withouthyper-parameteradjustment,itcanprovideafair
forecast.
4) It overcomes the problem of decision tree overfitting.
5) At the node's splitting point in every randomforesttree,a
subset of characteristics is chosen at random.
Fig. 4.2 Comparison of Accuracy of Unbalanced Data
Fig. 4.3 Comparison of Accuracy of Balanced Data using
Smote
Fig. 4.4 Comparison of Accuracy of Balanced Data using
Smote ENN
5.5 Pickle
For serializing and de-serializing a Python object structure,
the Python pickle package is utilized. Pickling an object in
Python allows it to be stored on a disc [3]. Pickle works by
first "serializing" the item before writing it to file. Pickling is
a Python function that converts a list, dict, or other Python
object into a character stream. The assumption is that this
character stream providesall ofthedata requiredtorecreate
the object in another Python function.

Fig. 4.5 Parameters for Evaluation
After applying these algorithms, we were able an achieve an
accuracy of 92 % which was given by Random Forest using
Smote ENN method after which we used the pickle library to
store the model and flask to integrate it with our webpage.
5. WEB TECHNOLOGIES
For our front-end part, we have used a Python framework
called Flask which is used to make a webpage. For our
project, we integrated our ML model into a website which is
used to get rich output and visualization.
5.1 Flask:
Flask is a Python-based web application framework. The
Werkzeug WSGI toolkit and the Jinja2 template engine are
the foundations of Flask. Both are Pocco initiatives.
And for the database part, we have used MySQL and also we
used the session to authenticate our login system with bank
users.
5.2 MYSQL:
MySQL is a quick, easy-to-use relational database
management system (RDBMS) that is utilizedbymanysmall
and large enterprises. MySQL AB, a Swedish business, is
responsible for its development, marketing, and support.
For our ML learning model, we used the Pickle python
library for integration with the flask webpage also we show
the graphs generated from the ML model into our webpage
for Easy to understand our users.
Normally Flask is used HTML templates to render the
webpage and for CSS we have used bootstrap and tailwind
CSS for validation JS function. Also, in the footer, we mention
our contributors’ details. Total 4 webpages are there on our
website which is Home, Login, Register, and Predict.
Screens on our website:
Home Page:
On this page, we make a basic webpage which is consist of
Navbar, Body, and Footer.
Fig. 5.1 Home Page
Login Page:
On this page for successful login, users have to give their
perfect credentials as per the registration. After successful
login bank users can use predict calculator or ML model for
their client/customers.
Fig. 5.2 Login Page
Register Page:
On this page, we ask for Basic details of usersforregistration
with the database.
Fig. 5.3 Registration Page
Predict Page:
On this page we made one predicting calculator field for our
ML model.
After filling all input fields we give the output as per the ML
model results and Accuracy.

Fig. 5.4 Predict Page
6. FUTURE SCOPE
Since the domain of the project (LAPD)revolvesarounddata
science and machine learning, it is possible to add Natural
Language Processing (NLP) based chat-bot which would
make the project more interactive and would help the user
in navigating the website or would recommend or resolve
queries regarding the usage of the application. The chat-bot
can be integrated either by using python language or by
using the Dialogflow platform powered by Google. In our
project (LAPD) we have developed a full-fledged websitefor
the end-user to predict the loan default. An alternativetothe
website can be a mobile application which would prove
much handier for the user. Cross-Platform technology
frameworks such as Flutter which is also backed by Google
can be brought into consideration for developing the
application [4]. Apart from this, currently, the dataset which
we have used is restricted to a particular organization.Ifit is
possible to collect a wider range of data,itwouldnotonlyact
as a knowledge base for the model but also help themodel in
understanding and predicting the loan default correctly
indirectly resulting in getting better accuracy
7. CONCLUSION
Due to a sudden halt caused by the COVID-19 pandemic, it
has affected almost every possible business of industries
thus leading to a financial dip. Many financial and banking
firms both large scale and small scale are the prime sectors
where the pandemic has largely impacted. The loan
defaulters rate rose high because of the lockdown and not
many were able to pay back. Hence, this invoked the need of
developing a system for predicting the loan defaults thereby
strengthening the loan sanctioning process.
This LAPD is thus an attempt to design and develop
a credit risk scoring model which could analyze and predict
the possibility of the loan default and thus reduce the
rejection rate of the new loan applicants and capital loss
faced by the bank which they would have made by
sanctioning the loan to the clients
REFERENCES
[1] Ahmad Al-qerem,GhaziAl-Naymat,MaysAlhasan,“Loan
Default Prediction Model Improvement through
Comprehensive Preprocessing and Features Selection”,
published.
[2] P. Maheswari, CH. V. Narayana, “Predictions of Loan
Defaulter - A Data Science Perspective”, published.
[3] Lin Zhua, Dafeng Qiua , Daji Ergua, Cai Yinga, Kuiyi Liub,
“A study on predicting loan defaultbasedonthe random
forest algorithm”, published.
[4] Hafiz Ilyas Tariq Aziz, Asim Sohail, Uzair Aslam, “Loan
Default Prediction Model UsingSample,Explore,Modify,
Model, and Assess (SEMMA)”, published.
[5] Mehul Madaan, Aniket Kumar, Chirag Keshri, Rachna
Jain and Preeti Nagrath, “Loan default prediction using
decision trees and random forest:Acomparativestudy”,
published.
[6] Harish Puvvada, VamsiMohanRamineedi,“LoanDefault
Prediction”, published.
[7] Haotian Chen, Ziyuan Chen, Tianyu Xiang, Yang Zhou,
“Data Mining on Loan Default Prediction”, published.

Loan Analysis Predicting Defaulters

More Related Content

Similar to Loan Analysis Predicting Defaulters (20)

More from IRJET Journal (20)

Recently uploaded (20)

Loan Analysis Predicting Defaulters