Volume 6, Issue 5, May – 2021 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Credit Card Score Prediction Using

Machine Learning
Simon Williams
M. Sc IT (Artificial Intelligence) Department of Information Technology University of Mumbai, India.

Abstract:- This study used a bank loan database to MACHINE LEARNING

check the applicability of the borrower classification The area of research includes a wide variety of fields
model and examined machine learning techniques. I including computer science, mathematics and information
developed auxiliary vector machine models, decision theory, cognitive psychological studies and philosophy (AI),
trees, and random forests, and compared their McCarthy et al1955. (Cook & Holder, 2001). Intelligent
prediction accuracy with benchmarks based on logistic biological systems are often regarded as prerogative to
regression models. They analyzed the performance solutions. Due to the multidisciplinary characteristics of
indicators based on the overall ranking. My results show artificial intelligence systems, they reflect multiple
that the performance of Random Forest is better than technologies and development strategies (Mira, 2008): (ii)
other models. In addition, the performance of the formal introduction of these models, (iii) implementation of
support vector machine model is poor when using linear programming and hardware strategies for those models.
and non-linear kernels. My results show that banks have
the opportunity to create value. Improve standard AI-based system is essential for adapting to changes in
predictive models by researching machine learning the environment. The AI discipline for automatic adaptation
techniques. programmes and data based learning is machine learning. Or
experience before (Mitchell, 1997). To this end, an
Keywords: Machine Learning, Artificial Intelligence, algorithm can be defined defining the sequence of
Supervised learning, Classification, Regression, Tensorflow. instructions converting input to output (Alpaydin, 2004).
Machine learning uses algorithms to differentiate important
I. INTRODUCTION patterns from irrelevant data patterns. For example machine
learning applications involve accurate medical diagnosis
Credit Scoring (e.g. breast cancer), reliable environmental disaster
The term "credit rating" is used to describe how the monitoring (e.g. forest fires) and sensory manufacturing
customer's default in financial obligations is assessed (Hand process monitoring (e.g. mechanics).
& Henley, 1997). The aim is to divide customers into two
groups: good and bad. Your financial obligations are Machine learning is mainly about supervised learning,
thought to be reimbursable. Members of unscrupulous and the aim is to learn how input and output are displayed.
groups are believed to be unable to fulfil their financial The data input is data describing a large number of single
obligations. objects, usually called examples or instances. The result is
the supervisor's result or result. Classification is a
A credit card is the simplest form of a series of supervised learning form in which different classes are
features used to assign a credit rating to customers that separated by a mapping (or discriminating function). There
shows their risks. The rating can be compared to the loan are several classes in the results. The results are called class
decision threshold. Because credibility is basically a names in machine learning. A classifier is called the
problem of discrimination (good or bad). The scoring is discriminating function. The training set is called a series of
traditionally divided into two main types according to the cases with known class names. The model is defined by
tasks and data used (Bijak and Thomas, 2012). Firstly, the many parameters during classification that are optimised for
level of application used in the loan application process to mapping instances to new invisible names of the instance.
evaluate the candidate's qualifications. Requesting. The data
used for this task are based on the repayment rate of the One-class classification (OCC) is an appreciation-
customer and, at a later date, the good-/bad situation. Banks based method that relies on a single example category to
must precisely predict the possibility of customer defaults in determine the normal or expected behavior of the target
different periods in order to be profitable (1 month, 3 category. This is a semi-managed form of classification
months, 6 months, etc.). High default risk customers can be because the training data contains examples that are only
banned to allow the bank to take adequate measures to labeled for the target category. This is in contrast to the
protect or control itself against loss. standard-driven classification method, which uses a
discriminative-based method to distinguish examples of
different categories. Various real-world problems, such as
machine failure detection (Sarmiento et al., 2005), fraud

IJISRT21MAY630 www.ijisrt.com 663

Volume 6, Issue 5, May – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
detection (Juszczak et al., 2008), and authentication
(Hempstalk, 2009).

Machine learning is a system that can learn from

examples through self-improvement without the need for
explicit programming by the programmer. This
advancement is based on the idea that machines can learn
from data in unique ways (for example, through examples)
to obtain accurate results. Combine data with statistical tools
to predict results. The company used this discovery to draw
practical conclusions. Machine learning is closely related to For example, a machine tries to understand the
data mining and Bayesian predictive modelling. The relationship between a person's salary and the possibility of
machine receives data as input and uses algorithms to going to a high-end restaurant, and it turns out that the
formulate a response. It is recommended to use typical machine finds a positive correlation between a person's
machine learning tasks. For users with a Netflix account, salary and going to a high-end restaurant: this is a model.
any movie or TV show recommendations are based on the
user's historical data. Technology companies are using Inferring
unsupervised learning to enhance the user experience After building the model, you can see how it performs
through personalized guidance. Machine learning is also on previously invisible data. Convert the new data into
used for various tasks, such as fraud detection, preventive feature vectors, refine the model and make predictions.
maintenance, asset portfolio optimization, task automation, training. No need to update rules or retrain the model. You
etc. Traditional programming is very different from machine can use previously trained models to draw conclusions about
learning. In traditional programming, programmers encode new data.
all rules after consulting with experts in the industry for
which software is being developed. Each rule is based on a
reason; after the logic is declared, the machine executes the
output. As the system becomes more complex, you need to
write more rules. During operation, this can quickly become
unstable.

DAT RUL
COMPUT
A ES
OUTP ER
The life of a machine learning programme, summed up
UT as follows
Machine  Define the problem,
How does Machine learningLearning
work?  Data gathering,
The brain in which all learning takes place is machine  Data display,
learning. Like humans, machines learn. People learn from  Algorithm training,
their own experience. The more we know, the easier it is to  Algorithm testing,
predict. For example, when we face an unknown situation.  Feedback collection,
The success rate is lower than the known situation. The way  Algorithm optimization.
of machine learning is the same. In order to make an
accurate prediction, the machine will consider an example.
 Cycle 4-7 to a satisfactory outcome.
If we give the machine a similar example, it can calculate  Use the predictive model.
the result. However, just like humans, it is difficult for
machines to make predictions when you provide examples Once the algorithm is successful, the learned
that have never been seen before. Learning and reasoning is knowledge is applied to a new data set.
the main objective of machine learning. The machine first
learns by recognising patterns. Data are attributed to this Machine learning Algorithms and where they are used?
discovery. As a data scientist, selecting carefully the data
you wish to supply to the computer is the most important
thing. List of problem-solving attributes. This is known as a
vector of function. Feature vectors can be considered a
subset of data for solving the problem. In order to simplify
reality and translate it into a model, the machine uses
complex algorithms. The training phase is therefore used to
describe and consolidate the data into the model.

IJISRT21MAY630 www.ijisrt.com 664

Volume 6, Issue 5, May – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Ø Regression problem

Classification
Assume you want to anticipate a customer's gender for
a salesperson. They'll take your client database's height,
weight, employment, salary, shopping cart, and other
information. Each customer's gender is known to you. It has
to be either male or female. The classifier's goal is to assess
whether a person is more likely to be male or female (i.e.
tags) based on the information provided (i.e the features you
collect).

When the model learns to distinguish between men

and women, it may make predictions based on fresh
information.
For instance, suppose you've recently gotten fresh
information from an unknown customer and you're trying to
figure out if it's a male or a woman. If the classifier predicts
a male-to-female ratio of 70%, the computer can conclude
that this consumer is 70% male and 30% female.

Machine learning may be separated into two types of There might be two or more classes in a label. In the
learning objectives: supervised and unsupervised. There are case above, however, there are just two classes. When the
other different algorithms available. classifier wants to forecast an item, though, it has hundreds
of options (such as glass, table, shoes, etc.). A class is an
Supervised Learning. object).
The algorithm finds the association between a certain
input and a specified output using feedback and training Regression
data. Professionals, for example, can deduct marketing The issue is regression when the output is a continuous
costs. Forecasting sales is done with the use of weather value. Financial analysts, for example, may be required to
predictions. You can utilise supervised learning when the forecast the value of stocks depending on a variety of factors
outcome is known. New data can be predicted by the (such as stocks, past stock returns, and macroeconomic
programme. indexes). Errors are possible.
There are two types of supervised learning:
Ø Classification problem

Algorithm Description Type

Name

Linear Enables each feature to be correlated to the result to predict future values. Regression
regression

Logistic Linear regression extension used for classification tasks. The output variable 3 is binary Classification
regression instead of continuous (e.g. just black and white) (e.g., an infinite list of potential colors)

Decision tree Classification or regression model that separates data-feature values into branches at Regression
decision nodes (e.g., if a feature is a colour, each potential colour becomes a new branch) Classification
until a final decision output is determined.

Naive Bayes Classification or regression model that separates data-feature values into branches at Regression
decision nodes (e.g., if a feature is a colour, each potential colour becomes a new branch) Classification
until a final decision output is determined.

Support For classification, the Support Vector Machine, or SVM, is commonly employed. The Regression (not
vector SVM method finds a hyperplane that divides the classes in the most efficient way. It very common)

IJISRT21MAY630 www.ijisrt.com 665

Volume 6, Issue 5, May – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

machine works best when combined with a nonlinear solution. Classification

Random The technique is based on a decision tree, which greatly improves accuracy. Random Regression
forest forest builds numerous basic decision trees and choose one label to return using Classification
the'majority vote' approach. The final forecast for the classification job will be the one
with the most votes, while the final prediction for the regression job will be the average
forecast of all the trees.

AdaBoost A classification or regression strategy that assesses a variety of models depending on their Regression
accuracy in predicting the outcome before making a choice. Classification

Gradient- Gradient-boosting trees are a cutting-edge classification and regression method. It focuses Regression
boosting on the preceding trees' mistakes and attempts to fix them. Classification
trees

Unsupervised learning
The algorithm checks input in an uncontrolled learning without explicit variables of output (for example, checking customer
demographics to identify patterns). You can use this to classify the data or to identify patterns for your algorithm and to classify
the information.

Algorithm Description Type

K-means Puts data into groups (k), each of which contains data with a comparable set of Clustering
clustering attributes (as determined by the model, not in advance by humans)

Gaussian mixture A k-means clustering extension that allows for more flexibility in the size and structure Clustering
model of group clusters.

Hierarchical Forms a categorization system by splitting groups along a hierarchical tree. Clustering
clustering Can be utilised for customers that have a Cluster loyalty card.

Recommender This helps to identify relevant data to make recommendations. Clustering

system

PCA/T-SNE Usually used to reduce the data's dimensionality. The techniques decrease the amount Dimension
of features to 3 or 4 greatest variance vectors. Reduction

II. APPLICATION OF MACHINE LEARNING Finance Industry

Ø Machine learning is becoming more and more
Augmentation: popular in the financial industry. Banks mainly use machine
 Machine learning can help people solve daily personal or learning to find data patterns and prevent fraud.
business problems without having to fully control the
results. Machine learning is utilised in a variety of Government organization
applications, including virtual assistants, data analysis, Ø The government uses machine learning to manage
and software solutions. Reduce human error. public safety and government services. Taking China as an
example, he has a big face. The government uses artificial
Automation: intelligence to prevent troublemakers from appearin.
 Machine learning can work completely autonomously in
each field without human intervention, such as B. Robots Healthcare industry
that perform important steps in production facilities. Ø Healthcare is one of the first industries to use
machine learning together with image recognition.

IJISRT21MAY630 www.ijisrt.com 666

Volume 6, Issue 5, May – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Marketing • Deep Q network
Ø Because of its enormous data access, artificial • State-Action-Reward-State-Action (SARSA)
intelligence is increasingly employed in marketing. • Deep Deterministic Policy Gradient (DDPG)
Researchers employed powerful mathematical methods such
as Bayesian analysis to estimate consumer benefits before Applications/ Examples of deep learning applications
the era of big data. The marketing department is relying on AI in Finance: The fintech industry is already using
AI to optimise customer interactions and marketing artificial intelligence to save time, reduce costs and increase
operations as data develops. value. Deep learning changes loans by improving
creditworthiness. Use artificial intelligence to better assess
Examples of the application of machine learning in the risks to assess the types and skills of job applicants.
supply chain. Underwrite is a financial technology company that provides
Machine learning produces great visual pattern artificial intelligence solutions to lenders. His method is
recognition results and opens up a slew of new physical fundamentally superior to traditional methods.
inspection and maintenance applications across the supply AI in HR: Underwear Armour is a sportswear company that
chain. Unsupervised learning allows you to quickly find is modernizing and modernizing its recruitment experience
comparable patterns in different data sets. During through artificial innovation. In fact, Under Armour reduced
transportation, the machine can perform quality inspections its retail store order time by 35%. The popularity of Under
on the damage and wear of the entire logistics centre. For Armour increased in 2012. An average of 30,000 resumes
example, the IBM Watson platform can detect damage to per month. Reading all these apps and starting the selection
shipping containers. and interview process took too long. The long process of
recruiting and matching employees affected Under Armours'
Watson integrates vision and system data for real-time ability to fully staff its retail stores. At the time, Under
tracking, reporting and recommendations. In the past year, Armour had all the necessary HR technologies, such as B.
reserve managers have relied heavily on the main methods Transactional solutions for procurement, implementation,
of estimating and forecasting reserves. By combining big tracking and onboarding, but these tools were not useful
data and machine learning, more advanced prediction enough. Under Armor, choose HireVue (an artificial
methods are introduced (20% to 30% higher than traditional intelligence provider for human resource issues) for on-
prediction tools). Indicates that it has increased by 2-3% due demand and real-time interviews. The result is bluffing.
to possible reduction in storage costs. They were able to reduce filling time by 35%, thereby hiring
more qualified employees.
Example of machine learning in Google Car. AI in Marketing: Artificial intelligence is a valuable tool
For example, everyone knows Google’s car. The car’s for managing customer service and personalization issues.
roof is full of lasers, which tell the car’s position relative to By using artificial intelligence technology to improve voice
the environment. There is a radar in the front, which can recognition and call routing in call center management, it
inform the car of the speed and movement of all surrounding can bring customers a smoother experience. For example,
cars. With all this data, not only can you figure out how to deep learning audio analysis. Enable the system to measure
drive a car, but you can also find and predict how a potential the mood tone of customers. If the client does not respond
driver will drive a car. The machine processes nearly 1 GB well to the AI chatbot, the system can route the conversation
of data per second. to a real human operator, who will be responsible for
solving the problem. In addition to the previous three
Deep learning examples, artificial intelligence is also widely used in other
Deep learning is a sort of computer software that industries.
simulates the brain's network of neurons. Because deep
neural networks are used, this is a subset of machine Artificial Intelligence
learning known as deep learning. The data is checked at
several stages by the machine. The number of layers in the
model represents the depth of the model. Deep learning is a
relatively recent term in the artificial intelligence field. The
learning step of deep learning is carried out via a neural
network.
Machine Learning ML Deep Learning
Reinforcement Learning
Reinforcement learning is a subset of machine learning
that trains the system by receiving virtual "rewards" or
"punishments" (essentially trial and error). Google
DeepMind used reinforcement learning to defeat the human
champion in the game of Go. It also improves gameplay by
providing smarter robots in video games.

One of the most famous algorithms:

• Q-learning

IJISRT21MAY630 www.ijisrt.com 667

Volume 6, Issue 5, May – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Difference between Machine Learning and Deep Learning.

Machine Learning Deep Learning

Data On a small/medium dataset, excellent results were Excellent performance with large amount of data.
Dependencies achieved.

Hardware Work on a low-cost computer. DL requires a powerful processor, preferably one

dependencies with a GPU, because it conducts a lot of matrix
multiplication.

Feature The characteristics that represent the data must be There's no need to figure out which feature best
engineering understood. describes the data.

Execution time From few minutes to hours Several weeks are possible. A large number of
weights must be computed using a neural
network.

Interpretability Some algorithms are simple to understand (logistic, Difficult to impossible

decision tree), while others are nearly hard to
comprehend (SVM, XGBoost)

Compared with deep learning, machine learning requires less data to train algorithms. Deep learning requires a large number
of diverse data sets to define the infrastructure. In addition, machine learning also provides faster model training. Exercise from a
few days to a week. The advantage of deep learning over machine learning is that it is very accurate. It is not necessary to know
which characteristics the data best represents. Neural networks have learned to highlight key characteristics. With machine
learning, you need to choose the features to include in the model.

TensorFlow Different groups use machine learning:

Google's TensorFlow is the world's famous deep • Researchers
learning library. The products used to improve search • Data scientists
engines, translations, headlines or recommendations in all • Programmers.
products by Google machine learning. To give a particular
example: We work faster and faster for Google users. Use Everyone can work together using the same
search enhancement AI. Google will give you suggestions instruments to improve its efficiency. Google doesn't just
for the next word when the user inserts a keyword in the have information. You have the world's largest computer, so
search bar. Google hopes that its extensive data sets are used TensorFlow is largely built. TensorFlow is a Google Brain
by machines and users can enhance their user experience. team library designed to speed up machine training and the
exploration of the deep neural networks. It can be used with

IJISRT21MAY630 www.ijisrt.com 668

Volume 6, Issue 5, May – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
several CPUs or GPUs, including mobile systems. It has (iv) Ratio: The characteristic value has the attribute of the
multiple conteners, such as Python, C++ or Java, which use interval data type, but has an absolute zero point (that
different languages. is, no negative value). For instance, Income to
expenditure. It is a collection of all possible outcomes.
TensorFlow Architecture The value training set S in the output space consists of n
Tensorflow architecture works in three parts: tuples (or instances).
• Preprocessing the data
• Build the model S = {(x1, y1),. , (xi,yi ),. , (xn, yn)} (1)
• Train and estimate the model
It is important to note that assuming that the examples
Tensorflow gets its name from the fact that it takes of x in S are independent and the same as (i.) of X, X is an
input in the form of a multidimensional array (also called a unknown but fixed common probability distribution
tensor). You can use this information to create a flowchart function P (x, y).
(also known as a chart) to run. Then proceed through this
versatile system to the other end and exit as an exit. Tensors THE LEARNING PROBLEM
enter, iterate over the list of operations, and then exit on the Using the training set S, the goal of supervised
other side, which is why it's named TensorFlow. learning is to approximate a function h : X → Y, and the
example xi can be mapped to its output value yi. A learning
It can be trained on several machines, then executed algorithm commonly called an inductor is used to perform
on an alternative machine, after the trained model has been the mapping function. The sensors of a particular training
created. set are called classifiers (Rokach, 2010). The space H of
Both GPUs and CPUs can be utilised to train and run classifiers or functions is called the space of classifiers or
the model. GPUs were created with video games in mind. the space of hypotheses. According to the result of Y, the
Stanford researchers discovered in late 2010 that GPUs are type of learning problem is usually defined as: (i)
also very good at matrix operations and algebra, making Regression learning, where Y = R; (ii) Learning
them incredibly quick for these types of tasks. A lot of classification, where Y = C, making C represent a set of
matrix multiplication is used in deep learning. Because categories, where C = {c1,.,cn}. The focus of this article is
TensorFlow is written in C++, it is extremely quick at the latter, that is, classification research. It is worth noting
computing matrix multiplication. TensorFlow, despite being that the general task of learning how to classify multiple
written in C++, can be accessed and controlled using various categories can be decomposed into many binary
languages, most notably Python. classification problems (Xu & Chan, 2003). In the binary
classification problem, the two categories can be represented
Finally, the TensorBoard is an important feature of as 0 and +1 respectively. For example, the type of borrower
TensorFlow. The TensorBoard lets you monitor what in the credit rating (i.e. Error) can be expressed as Y = {0,
TensorFlow does graphically and visually. +1}.

III. MAIN CONCEPTS MINIMIZING RISK

In order to select the best classifier for the hypothesis
This section provides a formal expression of the space, the loss function is used as a quantitative measure of
elements involved in supervised learning. In a typical the corresponding relationship between the predicted h(x)
supervised learning environment, a training set S of example and the expected result y. The best function h is the
x ∈ X and related output values y ∈ Y are provided. X is the minimum expected error (risk),
set of all possible examples in the input space, where X =
{x1,. , xi,. , xn}. Usually each example of x describe with a
vector of features or attribute values. Generally, in machine R(h) = L(h(x),y) = L(h(x),y) dP(x,y)
learning text, a function can be regarded as one of two data (2)
types:
(i) number: the feature value is a real number; or Where L represents the correspondingly selected loss
function. For binary classification, the loss function is
(ii) categorical: the feature value is a member of a usually 0/1 loss. If y = h(x), then L(h(x), y) is 0, otherwise it
predetermined finite set.
is 1. (X, y) is unknown, and the risk cannot be minimized
directly, but the solution is close to the expected minimum
The statistical text is different in the expansion of data
error of the available training set S. and is discriminatory.
types, including:
Classification basis (Cunningham et al.) The generation-
(i) Nominal value: the characteristic value is a member of based method checks common P(x,y) or P(y|x)P(x)
an unordered set, such as {tenant, owner, other}; probability models, and uses Bayes' theorem to obtain the
(ii) ordinal: the characteristic value is ordered The members required posterior probability.
of the set S. {high, medium, low};
(iii)Interval: the characteristic value is measured in a fixed
and equal unit, and is a member of an ordered set, for
example The temperature is in degrees Fahrenheit;
(3)

IJISRT21MAY630 www.ijisrt.com 669

Volume 6, Issue 5, May – 2021 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Thus the sum in the denominator is taken over to all review." Journal of the Royal Statistical Society:
categories. Discriminant method research assigns the input Series A (Statistics in Society) 160.3 (1997): 523-541.
attribute x directly to the class name y, that is, the back of P [6]. Sabeti, Behnam, et al. “Credit Risk Rating Using State
(y | x). Machines and Machine Learning.” International
Journal of Trade, Economics and Finance, vol. 11, no.
IV. SUMMARY 6, 2020, pp. 163–168.
[7]. Wallis, Mark, et al. “Credit Rating Forecasting Using
Banks use credit ratings to classify and assess the risk Machine Learning Techniques.” Advances in Data
of loss due to actual or perceived changes in customer Mining and Database Management Managerial
capabilities and willingness to pay financial liabilities. Perspectives on Intelligent Big Data Analytics, 2019,
pp. 180–198.
Initially, the credibility of private clients was taken as [8]. Tsai, Chih-Fong, and Ming-Lun Chen. “Credit Rating
a subjective measure. by Hybrid Machine Learning Techniques.” Applied
Soft Computing, vol. 10, no. 2, 2010, pp. 374–380.
The normal operation and improvement of the rating [9]. Weston, Liz Pulliam. Your Credit Score: How to
system is an obvious subject of concern to banks, customers Improve the 3-Digit Number That Shapes Your
and regulators, such as the latest developments in the world Financial Future. FT Press, 2016.
economy. [10]. Understanding Your Credit Report and Credit Score.
Financial Consumer Agency of Canada, 2012.
A scorecard is a digital scale used to assign levels to
customer characteristics to obtain a value that represents the
risk that the customer is considering you have not fulfilled
your financial obligations to other customers.

The scorecard development process is divided into

three main stages: (i) creating a data set; (ii) modeling; (iii)
documentation. This introduces various methods and
methods used in the creation and modeling of data sets. The
development of a scorecard is a detailed process and many
aspects need to be considered. Demographic trends and
economic events can cause various situations. Which
standards and accepted dashboard design methods are not
suitable.

One of these problems arises when the sample

contains few default values, which makes it difficult to
construct a reliable scorecard. After all, due to the
sensitivity of privacy laws and companies, obtaining actual
credit information is a challenge for many scholars. With the
help of artificial data, scientists can overcome these
limitations and create special conditions for studying
specific problems.

REFERENCES

[1]. Li, Xiao-Lin, and Yu Zhong. "An overview of

personal credit scoring: techniques and future work."
(2012).
[2]. West, Jarrod, and Maumita Bhattacharya. "Some
Experimental Issues in Financial Fraud Mining."
Procedia Computer Science 80 (2016): 1734-1744.
[3]. Yap, Bee Wah, Seng Huat Ong, and Nor Huselina
Mohamed Husain. "Using data mining to improve
assessment of creditworthiness via credit scoring
models." Expert Systems with Applications 38.10
(2011): 13274-13283.
[4]. Jayagopal, B. "Applying Data Mining Techniques to
Credit
[5]. Hand, David J., and William E. Henley. "Statistical
classification methods in consumer credit scoring: a

IJISRT21MAY630 www.ijisrt.com 670