See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/361293278

Development Of Credit Scoring Model For Borrowers Using Machine Learning

Techniques

Article in PERSPEKTIF · June 2022

DOI: 10.31289/perspektif.v11i3.7180

CITATIONS READS

4 632

3 authors, including:

Rebecca Etim Ekong Kuboye Bamidele Moses

1 PUBLICATION 4 CITATIONS
Federal University of Technology
36 PUBLICATIONS 172 CITATIONS
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Kuboye Bamidele Moses on 28 June 2022.

The user has requested enhancement of the downloaded file.

 PERSPEKTIF, 11 (3) (2022): 829-838
DOI: 10.31289/perspektif.v11i3.7180

PERSPEKTIF
Available online http://ojs.uma.ac.id/index.php/perspektif

Development of Credit Scoring Model for Borrowers Using

Machine Learning Techniques
Rebecca Etim Ekong1), Kolawole Gabriel Akintola2), &
Bamidele Moses Kuboye3)

1)Computer Science Department the Federal University of Technology Akure, Ondo State,
Nigeria
2)Software Engineering Department, the Federal University of Technology Akure, Ondo State,

Nigeria
3)Information Technology Department the Federal University of Technology Akure, Ondo

State, Nigeria

Received: April 20, 2022; Reviewed: April 20, 2022; Accepted: May 26, 2022

Abstract

Financial organizations such as banks have experienced an increase in demand for loans from borrowers over the
years. These organizations are highly interested in knowing whether a borrower can pay back if granted the loan
requested. Granting loans to defaulters can cripple the business, hence, these financial organizations are compelled
to evaluate credit worthiness of clients using the credit history of borrowers. Credit scoring is a technique used in
predicting the probability that a borrower will default. Several techniques have been adopted over the years such as
statistical and machine learning techniques, however, Machine learning techniques have been found to perform
better than the statistical techniques because they solve the challenges faced by credit analyst by automating the
processing and extraction of knowledge from data. The objective of this work is to improve upon the Artificial Neural
Network machine learning technique by adopting a better feature extraction technique. The methodology adopted
in this research is to use Rough Set Theory (RST) for relevant and efficient feature selection and Multi-Layer
Perceptron (MLP) Neural Network for classification. To test the models, the Australian and the German credit
datasets were used in the Anaconda machine learning platform. The results obtained from the research was
compared with some other machine learning models such as: Support Vector Machine, Random Forest, Decision
Tree, Logistic Regression, Naive Bayes, K-Nearest Neighbour and Artificial Neural Network using standard
evaluation metric to ascertain its performance on the two datasets. The results show that this work outperforms all
other models in any of the metrics considered. This research therefore has been able to show that the model is good
for credit scoring and has improved performance.

Keywords: Classification Techniques; Credit Scoring; Machine Learning Techniques; Rough Set Theory; Multi-Layer
Perceptron.

How to Cite: Ekong, R.E. Akintola, K.G. & Kuboye, B.M. (2022). Development of Credit Scoring Model for Borrowers
Using Machine Learning Techniques. PERSPEKTIF, 11 (3): 829-838

*Corresponding author: ISSN 2085-0328 (Print)

E-mail: ekongrebecca@gmail.com ISSN 2541-5913 (online)

829
 Ekong R.E., Akintola K.G. & Kuboye, B.M, Development of Credit Scoring Model for Borrowers Using
Machine Learning Techniques

INTRODUCTION automating the processing and extraction of

Due to inadequacy of resources, the knowledge from data. The size of data has
concept of credit became paramount for increased tremendously over the years which
individuals, families, governments, and make it suitable for using machine learning
businesses in the realization of their objectives. techniques to learn the data. However, these
The lending agencies assist borrowers to meet large datasets have much of noisy, irrelevant or
up expenses by providing loans to creditworthy misleading features which resulted in high
borrowers. The decision of granting loans is very computational cost (İlter, Kocadağlı &
critical and are analysed carefully to ascertain Ravishanker, 2019). In order to reduce the cost
whether these lending agencies can survive and of computation, feature extraction has become
make profit. The Subjective approach and necessary when dealing with large datasets.
Objective approach are the credit-granting There are several machine learning models
methodologies that financial organizations for classifying credit applicants, and choosing
might employ to assess credit applicants' the right one can enhance credit scoring
repayment potential (Bougard, 2017). The accuracy significantly (Hilscher & Wilson, 2017).
subjective approach uses judgment by the credit There is no existing model without
analyst to analyse a borrower’s character, shortcomings, therefore, the combination of
collateral and ability to repay (Pabuccu & Ayan, models can produce an improved credit score
2017). Also, recommendations from the (Nyoni & Matshisela, 2018). As regards this
employer of the borrower or previous lender can issue, six data machine learning algorithms are
be used to make a decision on who is entitled to selected. The goal is to use these machine
receive credit. Consistent choices and learning approaches on publicly available credit
inaccuracies made by various credit analysts for datasets from Germany and Australia to see
almost the same application plague the which categorization methodology produces the
subjective method. This strategy is also beset by best performance using standard evaluation
high training expenses, which arise when a metrics.
credit analyst is required to complete training The rest of this paper is structured as
before being able to approve an applicant. follows: Section 2 is the review of the related
Financial institutions have been motivated to literature. Section 3 subsequently describes our
explore objective methodologies and seek to approach in implementing the credit scoring
precisely anticipate the chance of default due to model. Section 4 describes the implementation
these flaws, rising demand for loans, and the procedure. Section 5 is the result and discussion
advancement of computer technology (Wu & from our findings, and the final section concludes
Pan, 2019). the study.
The urgency for a credit scoring model Credit scores have been calculated using a
became important because of the rapid growth variety of methods. It incorporates both
of industry which has caused an increase in the statistics and machine learning methodologies.
workload of the credit analysts. Individual Two researchers in 2020 carried out a review on
borrower’s credit worthiness should be analysing credit scoring using machine learning.
ascertained before credit is given to them They pointed out that the traditional method of
because failure to pay back can be very risky. credit scoring which is basically statistical has
Some financial institutions continue to use the become less popular due to the difficulties faced
traditional approach because of the by credit analysts and the current trend is the
unavailability of suitable data for machine use of machine learning to evaluate a credit score
learning techniques (Moradi & Mokhatab Rafiei, of an individual. Their analysis was derived from
2019). Also experts found it easier with the rules related works from previous researchers. The
they defined themselves rather than the machine result showed the efficacy of machine learning
learning models over which they had little or no over statistical models. The research gap was
control over. These were the two major reasons from their source of judgment which was drawn
why expert systems were preferred over solely from existing works. They did not carry
machine language. out any experiment analysis to validate their
Machine learning technique has ability to claim (Rudra Kumar & Kumar Gunjan, 2020).
solve these challenges faced by credit analyst by

830
 PERSPEKTIF, 11(3) (2022): 829-838

A paper published in 2017 showed the authors found out that one of the challenges of
suitability of Extreme Machine Learning (ELM) low accuracy was the presence of redundant
as a predictive model for credit scoring. The attributes in their credit dataset. They proposed
result of their research proved that ELM when a credit model to optimize the Decision Tree
compared to other models used in their research model. The result showed C5.0 was a better
produced a better result. However, only three option. The setbacks in their approach was that
metrics were used in evaluating the model they only used accuracy for performance
performances. In our research, six performance evaluation.
metrics was employed to investigate the Furthermore, previous researches used
suitability of our credit scoring model for Australian and German credit dataset to analyse
prediction (Bequé and Lessmann, 2017). credit risk (Thanawala, 2019; Pandey, Jagadev,
In a previous research published in 2018, Mohapatra & Dehuri, 2017). These authors
the researchers investigated how commercial focused only on accuracy as a measure to
banks could predict loans using machine ascertain which model did better. Although
learning techniques. They developed a credit classification accuracy is widely used by many
model using KNN classifiers. They discovered researchers because it is easy to understand and
that there was no perfect model from their compute, it is not a reliable measure when
previous research. The credit dataset was dealing with an imbalanced dataset. In this
analysed and their classifier was implemented research, the proposed model performance was
using R. The result of using a combination of evaluated using various metrics.
KNN classifiers produced an accuracy of 75%
(Arutjothi & Senthamarai (2018). The limitation RESEARCH METHODS
of their research is their accuracy result was low. The architectural diagram for our
An improved accuracy is necessary to minimize proposed credit scoring model for this research
the credit risk of financial institutions and is presented in Figure 1. This model was divided
maximize profit. into three parts namely, the data preparation,
In another publication in 2018, the authors the feature selection and the classification. In the
carried out a prediction analysis on borrowers data preparation part, data was collected from
defaulting using machine learning model and an online repository, cleaned from missing
deep learning model. The paper highlighted key values by replacing the missing value with
components in approving credits to borrowers corresponding mean or median of all instances
as feature selection algorithms, classification and the obtained samples were discretized using
models, evaluation metrics and credit analysts. Decision Tree model. Attribute reduction was
They observed that the tree models were more carried out in the feature selection part using
stable than deep learning models (Addo, Guegan, Rough Set Theory (RST) and then the result
& Hassani, 2018). obtained from performing feature selection was
Three authors in Zhang, Yang & Zhou, divided into training set and test set. The
(2018) also researched credit rating using three classification part is used to ascertain the
machine learning models for feature selection. creditworthiness of a borrower using Multi-
The models are C5.0, CHAID and CART. The layer Perceptron (MLP) Neural Network.

831
 Ekong R.E., Akintola K.G. & Kuboye, B.M, Development of Credit Scoring Model for Borrowers Using
Machine Learning Techniques

Figure 1. System architecture for the RST+MLP model

The dataset used in this paper are German has been used in most scientific papers available
and Australian credit dataset which was to predict credit score. The major features of
retrieved from the UCI Machine Learning datasets are mentioned in Table 1.
Repository (https://archive.ics.uci.edu/) which

Table 1. Features of the credit scoring datasets

Dataset No. of No. of Classes Nominal Attributes Numerical Attributes Good/Bad Credit %
Instances (Good & Bad)
German 1000 2 13 7 70/30
Australian 690 2 8 6 55.50/44.50

Once data was collected, the datasets were performing RST (Becker, Radomska-Zalas &
queried to check if there were any missing Ziemba, 2020), Indiscernibility, Reduct, and Core
values. There were 37 occurrences in the are the concepts used in our RST methodology.
Australian sample with one or more missing 𝑇 = (𝑈, 𝐴, 𝐶, 𝐷), where 𝑈 denotes the
variables. Replacement method was used in universe of discourse, 𝐴 denotes a collection of
replacing the missing value with corresponding basic features and 𝐶, 𝐷 ⊂ denotes two subsets
mean or median of all instances. If the attribute of features known as conditional and decision
is nominal, we used median but if numerical we attributes, respectively . Let a ∈ A, P ⊆ A. The
used the mean. Also, the numerical variables indiscernibility relation 𝐼𝑁𝐷(𝑃) is expressed as
contained in the credit dataset were transformed (Qu et al., 2020):
into nominal variables by discretization while
𝐼𝑁𝐷(𝑃) = {(𝑥, 𝑦) ∈ 𝑈 × 𝑈 ∶ ∀ 𝑎 ∈
the nominal variables were represented in
𝑃, 𝑎(𝑥) = 𝑎(𝑦)} (1)
numerical code to aid in easy interpretability of
the feature. Normalization is then performed to Where (𝑥, 𝑦) are the cases, while a(x)
obtain a range of values between 0 and 1. signifies the value of attribute 𝑎 for case 𝑥 , and
A large number of attributes in datasets (𝑥, 𝑦) denotes a pair of cases. Thus, if (𝑥, 𝑦) ∈
may be unnecessary. The computing time for 𝐼𝑁𝐷(𝑃), it implies that x and y are identical with
categorization will increase if these duplicate regard to P.
attributes are not deleted. As a result, feature In order to select the most relevant
selection is utilized to choose crucial features attributes, we used a selection strategy in setting
that are required while creating models. In this a threshold value (Chowdhury & Turin, 2020)
research, RST was used to perform feature and it is calculated from the priority values
selection. Although there are numerous concepts representing the significance of each feature.
in RST (Skowron & Dutta, 2018) and methods of

832
 PERSPEKTIF, 11(3) (2022): 829-838

𝑐𝑚𝑎𝑥 −𝑐𝑚𝑖𝑛
𝛽(𝑐) ≥ + (2) structure as the original dataset. There exist
3
Where 𝑐𝑚𝑎𝑥 and 𝑐𝑚𝑖𝑛 respectively reflect multiple reducts for a dataset. A set of features
the ranking vector's greatest and minimum 𝑅 ⊆ 𝐶 is called a reduct of 𝐶 . The set of all the
values. The goal of this technique is to provide a attributes indispensable in 𝐶 is denoted by
measure independent of the actual distribution 𝐶𝑂𝑅𝐸(𝐶).
of the priority values and to approximate the
selection of features having their priority values 𝐶𝑂𝑅𝐸(𝐶) = ⋂ 𝑅𝐸𝐷(𝐶) (3)
larger than the median value. The result is used
to form reduct, which is a minimal subset of Where 𝑅𝐸𝐷(𝐶) is the set of all reducts of 𝐶 .
attributes providing the same equivalence class

Attribute reduction pseudocode for RST

Input: 𝐶 - the set of all conditional attributes;
𝐷 - the set of decision attributes
t - total number of instances
Process:
RED(C) ← {}
P ← {}
Foreach c є C
Get the values of c denoted 𝑉𝑐
c ← 𝑉𝑐 /𝑡

4. cmax ← get maximum of C

5. cmin ← get minimum of C
6. P← all features with cmax
7. C ← C- P
8. Check if Indiscernibility in C does not exist using equation (1)
9. set a new threshold using equation (2) and remove all features above threshold
10. Add to P
11. then do (7) and (8) until indiscernibility exist
12. R← P
13. continue till all the minimal reducts are found
14. Use equation (3) to get the CORE
15. return CORE.
Output: CORE - the feature subset

Data Split. Before classification was done, layers were used and its derivation is expressed
the pre-processed dataset was divided into two in equation 4 to 6.
sections: a training dataset and a test dataset
with 70:30 as the ratio. The training dataset was 𝑛𝑘, 𝑡 = 𝜔𝑘,0 + ∑𝑖𝑖=0 𝜔𝑘, 𝑖 𝑥𝑖, 𝑡 (4)
utilized for classification, whereas the test
dataset was used to assess the classification 1
accuracy. 𝑁𝑘,𝑡 = 1+ 𝑒 −𝑛𝑘, 𝑡 (5)
Model Selection. The multi-layer
perceptron (MLP) model trained with the back Equation (4) and (5) shows how the nodes
propagation was employed in this study. are being calculated from the set of input x with
Let 𝑥 = (𝑥1, 𝑥2, … , 𝑥𝑛 ) as a set of input and their corresponding arbitrary weights and the
ℎ = (ℎ1 , … , ℎ𝑚 ) as the set of nodes in the result (nk,t) is passed into the activation function.
hidden layer. There can be more than one hidden The activation function used is sigmoid function.
layer in MLP (Bekesiene, Smaliukiene, &
Vaicaitiene, 2021). In this research, two hidden

833
 Ekong R.E., Akintola K.G. & Kuboye, B.M, Development of Credit Scoring Model for Borrowers Using
Machine Learning Techniques

Equation (6) is the result of the calculation this study. Each criterion has its strengths and
of all the nodes in the input and hidden layers weaknesses. In this study, it was found that using
while 𝛾0 is the bias. a combination of these indicators rather than a
single measure to evaluate the performance of
credit scoring models is preferable. The
𝑦𝑡 = 𝛾0 + ∑𝑙𝑙=1 𝛾𝑙 𝑃𝑘, 𝑖 + ∑𝑖𝑖=0 𝛽𝑥𝑖, 𝑡
confusion matrix is the source of all of these
(6) performance evaluation criteria. The confusion
matrix is a form of square matrix table used in
Model Performance Evaluation. showcasing the classification model result on
Accuracy, Precision, F1 Score, Recall, Type I test datasets. Confusion matrix gives an idea of
error and Type II error were among the what classification models predict correctly and
performance evaluation techniques employed in what they predict incorrectly.

Table 2. Confusion Matrix

Predicted Class
Non-Defaulter Defaulter
Actual Non-Defaulter TN FP
Class Defaulter FN TP

true class membership is good, TN is

Table 2 shows the structure of the observations classified as good whose true class
confusion matrix where TP is observations membership is good and FN is observations
classified as bad whose true class membership is classified as good whose true class membership
bad, FP is observations classified as bad whose is bad.

Table 3. Summary of how each performance criteria can be calculated

Metric Formula
Accuracy 𝑇𝑃 + 𝑇𝑁
𝑇𝑃 + 𝑇𝑁 + 𝐹𝑃 + 𝐹𝑁
Precision 𝑇𝑃
𝑇𝑃 + 𝐹𝑃
Recall 𝑇𝑃
𝑇𝑃 + 𝑇𝑁
F1 Score 2 ∗ 𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 ∗ 𝑟𝑒𝑐𝑎𝑙𝑙
𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 + 𝑟𝑒𝑐𝑎𝑙𝑙
Type I 𝐹𝑁
𝐹𝑁 + 𝑇𝑁
Type II 𝐹𝑃
𝐹𝑃 + 𝑇𝑃

RESULT & DISCUSSION existing credits at this bank. For the Australian
A total of 12 out of 20 conditional dataset, 7 out of 14 attributes were selected. The
attributes met the criteria when feature attributes are A2, A3, A5, A6, A7, A10 and A14.
selection was performed on the German credit Figure 2 represents the confusion matrix of
dataset. These features are Existing checking the model on the Australian dataset. A total of
account status; Month-long duration; History of 90% accuracy, 93% precision, 89% for recall and
credit; Bonds and savings accounts; Rate of 91% for f1 score were obtained. Figure 3 shows
installments expressed as a proportion of the confusion matrix of the model on the German
disposable income; Job; Other installment plans; dataset. A total of 87% for accuracy, 84% for
Other guarantors / debtors; property; current precision, 88% for f1 score and 91% for recall
residential address, age in years and Number of were obtained.

834
 PERSPEKTIF, 11(3) (2022): 829-838

Figure 2 Confusion Matrix obtained from the model on Australian dataset

Figure 3 Confusion Matrix obtained from the model on German dataset

In order to ascertain how well our model compared with the result obtained from a
performed, the model was benchmarked with 6 previous research by Aithal & Jathanna (2019)
other models on the Australian and German and it outperformed the other models as shown
credit dataset. The models used were K-Nearest in Table 4. Our feature extraction methodology
Neighbour (KNN), Logistic Regression (LR), produced a higher accuracy when applied to a
Naive Bayes (NB), Random Forest (RF), and larger dataset compared to the novel credit data
Multi-Linear Perceptron (MLP) Neural Network model (Zhang, et al., 2018). It is important to
Decision Tree (DST) and Support Vector note that the credit history of the borrowers was
Machine (SVM). It was observed that the not considered as an important feature. This
proposed model performed best in all the six might be misleading because it is important to
performance metrics. study a borrower’s history before such an
The proposed model has an accuracy of applicant is granted a loan.
90% and 87% for the Australian and German The error results equally reveal that the
credit dataset respectively. This shows that Type I and Type II errors are the lowest when
there is a significant improvement with a compared with all the models used. Table 5 and
minimum of 4% in terms of accuracy, precision, Table 6 shows the performance of the proposed
f1 score and recall for each of the dataset used model on Australian and German datasets
when compared to other machine learning respectively compared to some existing models.
techniques. Furthermore, MLP+RST was

835
 Ekong R.E., Akintola K.G. & Kuboye, B.M, Development of Credit Scoring Model for Borrowers Using
Machine Learning Techniques

From the tables, it is observed that the proposed

model outperformed all other models.

Table 4. Comparative analysis of performance from previous research

ACC(%) REC(%) PREC(%) F1-S(%) Error(%
)
LR 75 91 76 83 25
NB 77 78 91 84 23
SVM 76 81 80 83 24
RF 78 80 91 85 22
CART 77 93 78 84 23
NN 74 80 82 81 26
RST+MLP 87 91 84 88 13

Table 5. Performance of machine learning techniques on the Australian dataset

MODEL ACC PRECISION RECAL F1 TYPE I TYPE II
L
DST 84 83 85 84 15 17
LR 86 87 84 86 15 13
KNN 66 78 44 56 38 22
NB 77 90 61 73 29 10
RF 86 88 81 85 17 12
SVM 86 84 87 86 13 16
MLP 76 81 68 74 27 19
RST +MLP 90 93 89 91 13 07

Table 6. Performance of machine learning techniques on German dataset

MODEL ACC PRECISION RECAL F1 TYPE I TYPE II
L
DST 69 48 46 47 23 48
LR 76 70 33 45 23 30
KNN 68 46 36 40 26 46
NB 69 48 53 51 21 48
RF 72 55 32 41 25 45
SVM 74 70 23 35 26 30
MLP 74 72 34 46 25 28
RST +MLP 87 84 91 88 10 16

Figure 4. Performance Evaluation on Australian Dataset

836
 PERSPEKTIF, 11(3) (2022): 829-838

Figure 5 Performance Evaluation on German Dataset

CONCLUSION research studies should consider the use of other

MLP with RST was considered in this paper feature selection algorithms.
for credit scoring. The models were compared
with six machine learning models using the ACKNOWLEDGMENTS
Australian and German credit datasets. The aim The authors are thankful to the Conference
of this study was to improve credit scoring with Organising Committee of Covenant University
improved performance while saving the Conference on E-Business and E-Government in
computational costs. This has become necessary Nigeria (CUCEEN 2021) chaired by Dr.
due to the increase in size of data making it Ugochukwu D. Abasilim for providing this
suitable for using machine learning techniques platform for the publication. Also, Covenant
to learn the data. However, these large datasets University Centre for E-Business and E-
have an abundance of noisy, irrelevant or Government Research is highly appreciated for
misleading features. Processing these large granting the approval for the publication of this
datasets comes with high computational cost. article.
The results of these tests revealed that MLP
credit scoring model with rough set features REFFERENCES
surpassed all other models in terms of accuracy, Aragon, L. V. (n.d.). (2001) C om m unal V iolence in
precision, recall and f1 score on the two credit Poso Addo, P., Guegan, D., & Hassani, B. (2018).
datasets. The Australian and German credit Credit Risk Analysis Using Machine and Deep
Learning Models. Risks, 6(2), 38.
scoring models had the highest accuracy at 90%
https://doi.org/10.3390/risks6020038
and 87 % respectively using 70:30 splitting ratio Aithal, V., & Jathanna, R. D. (2019). Credit Risk
of the training and test dataset. Assessment using Machine Learning
MLP with RST is a better method in credit Techniques. International Journal of
scoring applications. To ascertain this, the Innovative Technology and Exploring
suitability of MLP+RST was established Engineering, 9(1), 3482–3486.
practically. Unlike previous studies that https://doi.org/10.35940/ijitee.a4936.11911
considered solely the predictive accuracy 9
(Sharifi et al., 2021; Siregar & Wanto, 2017), Arutjothi, G., & Senthamarai, C. (2018). Prediction of
other performance metrics were also considered loan status in commercial bank using machine
learning classifier,. In Proceedings of the
in this work such as precision, recall and the f1
International Conference on Intelligent
score. We tested MLP on all the features and also Sustainable Systems, ICISS 2017, 416–419.
on the reduced features captured by RST. In Becker, J., Radomska-Zalas, A., & Ziemba, P. (2020).
every performance metrics MLP+RST Rough set theory in the classification of loan
outperformed the other models considered in applications. Procedia Computer Science, 176,
this study. The finding also shows that MLP when 3235–3244.
fed with appropriate data shows a more https://doi.org/10.1016/j.procs.2020.09.125
established classifier in the application area of Bekesiene, S., Smaliukiene, R., & Vaicaitiene, R.
credit scoring. It is recommended that future (2021). Using Artificial Neural Networks in
Predicting the Level of Stress among Military

837
 Ekong R.E., Akintola K.G. & Kuboye, B.M, Development of Credit Scoring Model for Borrowers Using
Machine Learning Techniques

Conscripts. Mathematics, 9(6), 626. Qu, J., Bai, X., Gu, J., Taghizadeh-Hesary, F., & Lin, J.
https://doi.org/10.3390/math9060626 (2020). Assessment of Rough Set Theory in
Bequé, A., & Lessmann, S. (2017). Extreme learning Relation to Risks Regarding Hydraulic
machines for credit scoring: An empirical Engineering Investment
evaluation. Expert Systems with Decisions. Mathematics, 8(8), 1308.
Applications, 86, 42–53. https://doi.org/10.3390/math8081308
https://doi.org/10.1016/j.eswa.2017.05.050 Rudra Kumar, M., & Kumar Gunjan, V. (2020). Review
Bougard, D. A. (2017). Agricultural credit models: of Machine Learning models for Credit Scoring
identifying high risk Analysis. Ingeniería Solidaria, 16(1).
applications. Scholar.ufs.ac.za. https://doi.org/10.16925/2357-
https://scholar.ufs.ac.za/handle/11660/647 6014.2020.01.11
2?show=full Sharifi, P., Jain, V., Arab Poshtkohi, M., Seyyedi, E., &
Chowdhury, M. Z. I., & Turin, T. C. (2020). Variable Aghapour, V. (2021). Banks Credit Risk
selection strategies and its importance in Prediction with Optimized ANN Based on
clinical prediction modelling. Family Medicine Improved Owl Search
and Community Health, 8(1), e000262. Algorithm. Mathematical Problems in
https://doi.org/10.1136/fmch-2019-000262 Engineering, 2021, 1–10.
Hilscher, J., & Wilson, M. (2017). Credit Ratings and https://doi.org/10.1155/2021/8458501
Credit Risk: Is One Measure Simplilearn.com (2021). An Overview on Multilayer
Enough? Management Science, 63(10), 3414– Perceptron (MLP). [online] Available at:
3437. https://www.simplilearn.com/tutorials/dee
https://doi.org/10.1287/mnsc.2016.2514 p-learning-tutorial/multilayer-perceptron.
İlter, D., Kocadağlı, O., & Ravishanker, N. (2019). Siregar, S. P., & Wanto, A. (2017). Analysis of Artificial
Feature Selection Approaches for Machine Neural Network Accuracy Using
Learning Classifiers on Yearly Credit Scoring Backpropagation Algorithm In Predicting
Data. Recent Advances in Data Science and Process (Forecasting). IJISTECH
Business Analytics, 200–204. (International Journal of Information System
Moradi, S., & Mokhatab Rafiei, F. (2019). A dynamic & Technology), 1(1), 34.
credit risk assessment model with data mining https://doi.org/10.30645/ijistech.v1i1.4
techniques: evidence from Iranian banks. Skowron, A., & Dutta, S. (2018). Rough sets: past,
Financial Innovation, 5(1). present, and future. Natural
https://doi.org/10.1186/s40854-019-0121- Computing, 17(4), 855–876.
9. https://doi.org/10.1007/s11047-018-9700-
Nyoni, E. E. T., & Matshisela, N. (2018). Credit scoring 3
using machine learning algorithims. Thanawala, D. D. (2019). Credit Risk Analysis Using
Zimbabwe Journal of Science & Technology, e- Machine Learning and Neural Networks. Open
ISSN 2409-0360, 13, 26–34. Access Master's Report, Michigan
Pabuccu, H. & Ayan, T.Y. (2017) the Development of Technological University.
an Alternative Method for the Sovereign https://doi.org/10.37099/mtu.dc.etdr/856
Credit Rating System Based on Adaptive Wu, Y., & Pan, Y. (2021). Application Analysis of
Neuro-Fuzzy Inference System. American Credit Scoring of Financial Institutions Based
Journal of Operations Research, 7, 41-55. on Machine Learning Model. Complexity,
Pandey, T. N., Jagadev, A. K., Mohapatra, S. K., & 2021, 1–12.
Dehuri, S. (2017). Credit risk analysis using https://doi.org/10.1155/2021/9222617
machine learning classifiers. 2017 Zhang, X., Yang, Y., & Zhou, Z. (2018). A novel credit
International Conference on Energy, scoring model based on optimized random
Communication, Data Analytics and Soft forest. 2018 IEEE 8th Annual Computing and
Computing (ICECDS). Communication Workshop and Conference
https://doi.org/10.1109/icecds.2017.838976 (CCWC).
9 https://doi.org/10.1109/ccwc.2018.830170
7

838

View publication stats