International Journal of Computer Science and Information Security (IJCSIS),

Vol. 22, No. 3, June 2024

CREDIT SCORING SYSTEM

P. Srinivas Rao1, Syed Ismat Maalikah 2, Telapolu Poojitha 3, S. Anjani Kaustub 4, T. Lokesh 5
1,2,3,4,5
Sreyas Institute of Engineering and Technology 1 Assistant Professor (AIML)

Abstract:
In the realm of banking and financial institutions, mitigating the risk of loan defaulters is
paramount to ensure sustained profitability and stability. Traditional methods for assessing credit
risks often involve manual processes that are resource-intensive and may not always yield accurate
results. In this context, neural networks offer a promising alternative for credit scoring systems.
By leveraging advanced machine learning techniques, neural networks can efficiently analyse vast
amounts of data to predict the likelihood of loan default. This project explores the application of
neural networks in credit scoring, highlighting their potential to enhance accuracy and efficiency
in risk assessment. The project implements various neural network architectures and
methodologies utilized in credit scoring, along with their advantages and limitations. Furthermore,
it examines the challenges and opportunities associated with integrating neural networks into
existing credit risk management frameworks. Ultimately, this research contributes to the ongoing
discourse on leveraging artificial intelligence in the financial sector to make informed decisions
and mitigate credit risk effectively.
Keywords: Credit scoring, Neural networks, Loan default, Risk assessment, Machine learning,
financial institutions, Artificial intelligence, Credit risk management, Predictive modelling.

I. INTRODUCTION:
Accurate credit risk assessments and mitigation are critical in today's banking and financial
institutions landscape to ensure continuous profitability and stability. Credit risk, which is
characterised by a higher chance of loan default, poses a serious threat to financial institutions
globally. Credit risk evaluations have traditionally been performed manually, which can be
inaccurate and resource-intensive. But with the development of sophisticated machine learning
methods—neural networks in particular—there may be a chance to completely transform credit
scoring models.
Credit scoring systems serve as pivotal tools for financial institutions to evaluate the
creditworthiness of potential borrowers. These systems analyse various factors such as credit
history, income levels, debt-to-income ratios, and employment stability to predict the likelihood
of loan default. The accuracy of these predictions directly impacts the institution's profitability and
overall financial health. Traditional credit scoring methods, while effective to some extent, often
fall short in handling the complexity and volume of data available in today's financial landscape.
A branch of artificial intelligence called neural networks presents a strong challenge to
conventional credit rating techniques. Large volumes of data can be processed quickly by these
sophisticated algorithms, and they can also spot complex patterns that might be difficult to find

https://google.academia.edu/JournalofComputerScience 1 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

through manual examination. Financial organisations can improve the precision and effectiveness
of their credit risk assessment procedures by utilising neural networks.
The objectives of this project encompass a comprehensive exploration of the application of neural
networks in credit scoring systems, particularly focusing on the utilization of sequential models
and SMOTE techniques. The study aims to investigate various neural network architectures and
methodologies employed in credit scoring, with a specific emphasis on understanding how
sequential models and SMOTE techniques can be effectively integrated.
The integration of neural networks into credit scoring systems holds immense potential to
revolutionize the way financial institutions assess credit risk. By leveraging advanced machine
learning techniques, institutions can enhance the accuracy, efficiency, and scalability of their credit
risk assessment processes. This project contributes to the ongoing discourse on the adoption of
artificial intelligence in the financial sector, paving the way for more informed decision-making
and effective risk management strategies.

II. LITERATURE SURVEY:

A Deep Learning Approach for Credit Scoring Using Feature Embedded Transformer by Chongren
et al [1] A transformer is introduced into the field of credit scoring based on user online behavioral
data and develop an end-to-end feature embedded transformer (FE-Transformer) credit scoring
approach. The transformer uses deep neural network to analyse the credit scores.
Corporate Credit Ratings Based on Hierarchical Heterogeneous Graph Neural Networks by Bo-
Jing et al [11] These models are based on statistical learning, machine learning and deep learning
especially graph neural networks (GNNs)
A novel federated learning approach with knowledge transfer for credit scoring by Zhongyi et al
[3] A privacy-preserving credit scoring method is proposed. We propose the federated knowledge
transfer (FedKT) approach to efficiently exploit the generic knowledge and specific knowledge
from the global model. The proposed FedKT can avoid insufficient learning of the local models
due to class imbalance.
Emerging Trends in Deep Learning for Credit Scoring: A Review by Yoichi, this paper explains
how tabular datasets are converted into images for the application of a two-dimensional
convolutional neural network (CNN) and how “black box” models using local and global rule
extraction and rule-based methods are applied in credit scoring.
Neural network as a tool to solve the problem of credit scoring by Vladimir et al [5] The developed
information system should solve the problem of credit scoring, which is to determine the
creditworthiness of the client. Based on the information of this client, as well as on historical data
about other customers, the system determines whether the person is solvent.
Flexible loss functions for binary classification in gradient-boosted decision trees: An application
to credit scoring by Jonah et al [6] This paper presents a novel flexible loss function for binary

https://google.academia.edu/JournalofComputerScience 2 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

classification based on the popular XGBoost implementation of gradient-boosted decision trees

(GBDT). GBDTs such as XGBoost are relatively transparent in comparison to other machine
learning techniques, which is desirable for financial applications.
Machine Learning Approach for Credit Score Predictions by Tsholofelo et al [7] This paper
addresses the problem of managing the significant rise in requests for credit products that banking
and financial institutions face. The aim is to propose an adaptive, dynamic heterogeneous ensemble
credit model that integrates the XGBoost and Support Vector Machine models to improve the
accuracy and reliability of risk assessment credit scoring models.
Credit Score Prediction by Krishna et al [8] Credit scores are used by lenders, such as banks and
credit companies, to evaluate the risk of lending money or extending credit to an individual. A
model is also made using one of these algorithms which provide minimum error of provide correct
predictions based on the previous data available. The model is made using random forest classifier
algorithm which gave the highest accuracy, out of all the machine learning algorithm, up to
79.97%.
Credit-Risk Prediction Model Using Hybrid Deep—Machine-Learning Based Algorithms, Tamiru
et al [9] Credit-risk prediction is one of the challenging tasks in the banking industry. In this study,
a hybrid convolutional neural network—support vector machine/random forest/decision tree
(CNN—SVM/RF/DT) model has been proposed for efficient credit-risk prediction. Different
parameter values were considered and fine-tuned throughout testing to select appropriate
parameters. In accordance with the experimental findings, a fully connected CNN, and a hybrid
CNN with SVM, DT, and RF, respectively, achieved a prediction performance of 86.70%, 98.60%,
96.90%, and 95.50%. According to the results, our suggested hybrid method exceeds the fully
connected CNN in its ability to predict credit risk.
Analysis and visualization of financial data with power BI by Pedro. This paper focuses on the
exploration and analysis of the importance of Business Intelligence (BI) tools, specifically
Microsoft's Power BI. the importance of Power BI as a fundamental tool in the modern era of data-
driven business, highlighting its ability to adapt to diverse business needs and its contribution to
informed and strategic financial decision making.

Author Proposed Result Analysis Description Accuracy

Algorithms
Chongren Feature Based on the Uses end-to-end AUC: 0.72
Embedded graphs the AUC is feature embedded
Transformer predicted. transformer credit
scoring approach.
The transformer
uses deep neural
network to

https://google.academia.edu/JournalofComputerScience 3 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

analyse the credit

scores.
BoJing Graph neural HHGNN is Using hierarchical 97.5%
networks capable of heterogeneous
intelligently graph neural
interpreting networks which
financial data and can fully model
has inestimable the hierarchy of
application value corporate features
for reducing heterogeneity of
financial risk relationships
through corporate between
credit rating. corporations.
Zhongyi federated The proposed A privacy- 81%
knowledge FedKT can avoid preserving credit
transfer insufficient scoring method is
learning of the proposed
local models due
to class
imbalance.
Yoichi convolutional Black Box and “Black box” 88%
neural network CNN models models using
(CNN) using rule-based local and global
methods. rule extraction
Vladimir Neural networks Based on The developed 87%
historical data information
system should
solve the problem
of credit scoring,
Jonah Gradient-boosted Binary extensions 91%
decision trees classification decrease the
based on the misclassification
popular XGBoost costs associated
implementation
with incorrectly
classifying
positive cases as
negative when
the class
imbalance is not
severe, which
has a financial

https://google.academia.edu/JournalofComputerScience 4 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

benefit for credit

scoring.
Tsholofelo XGBoost and Ensemble credit Addresses the 83%
Support Vector model that problem of
Machine models integrates the managing the
XGBoost and significant rise
Support Vector
in requests for
Machine
credit products
that banking and
financial
institutions face.
Krishna Random forest The model is . A model is also 79.97%.
classifier made using made using one
random forest of these
classifier algorithms
algorithm which provide
minimum error
of provide
correct
predictions
based on the
previous data
available.
Tamiru Hybrid fully connected Convolutional 86.70%, 98.60%,
convolutional CNN and a hybrid neural network— 96.90%, and
neural network CNN with SVM, support vector 95.50%
DT, and RF machine/random
forest/decision
tree model has
been proposed for
efficient credit-
risk prediction

III. PROPOSED SYSTEM

https://google.academia.edu/JournalofComputerScience 5 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

Figure1: System Architecture of the proposed system

Dataset-
The dataset is taken from Kaggle and the dataset is clean and has no missing values and outliers.
Data Pre-processing –
As we have a clean dataset only encoding of certain features like Occupation, Payment of
Minimum Amount and few are done and saved
After encoding process , as the model has multi-class prediction the three classes need to have
equal number of class distribution which is achieved using SMOTE
Methodology –
The goal of the model is to predict if the credit score is good , bad or standard.
The neural network architecture used for credit score prediction comprises several layers, each
serving a specific purpose to enhance the model's performance and generalization ability.
Input Layer: The input layer accepts the feature vectors representing the input data, such as credit
mix, payment behavior, age, annual income, and number of bank accounts.
Hidden Layers: Dense Layers: The model consists of multiple dense (fully connected) layers, each
followed by an activation function (ReLU). Dense layers allow the model to learn complex patterns
and relationships within the data by connecting every neuron from the previous layer to every
neuron in the current layer.

https://google.academia.edu/JournalofComputerScience 6 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

Activation Function (ReLU): Rectified Linear Unit (ReLU) activation function is chosen for its
simplicity and effectiveness in handling non-linearities in the data. ReLU introduces non-linearity
to the model, enabling it to learn and represent complex relationships between features.
Normalization Layers (Batch Normalization):
Batch Normalization: Batch normalization layers are added after each dense layer to normalize the
activations of the previous layer. This helps in stabilizing and accelerating the training process by
reducing internal covariate shift. Batch normalization ensures that the inputs to each layer have a
mean close to zero and a standard deviation close to one, which prevents gradients from vanishing
or exploding during training.
Regularization (Dropout):
Dropout: Dropout layers are employed to prevent overfitting by randomly dropping out (setting to
zero) a fraction of the input units during training. This regularization technique introduces noise
to the model and forces it to learn more robust features, thus improving its generalization ability
on unseen data. In this model, dropout is applied after some dense layers to reduce the risk of
overfitting while retaining the model's ability to learn complex patterns.
Model Training:
Capturing Credit Patterns Over Time: The credit scoring process involves analysing various
financial and personal factors to assess an individual's creditworthiness. Training the neural
network model over epochs allows it to learn and adapt to the complex patterns inherent in credit
data. Each epoch (shown in figure 2,3) enables the model to refine its understanding of how factors
such as credit mix, payment behaviour, age, and annual income influence credit scores, ensuring
that the model captures nuanced relationships and dynamics over time.
Adaptation to Evolving Credit Trends: Credit markets and consumer behaviours are subject to
change due to economic fluctuations, regulatory updates, and shifts in lending practices. Training
the model over epochs ensures that it remains adaptable to evolving credit trends. By continually
exposing the model to updated data and retraining it over epochs, financial institutions can ensure
that their credit scoring systems reflect current market conditions and regulatory requirements,
enhancing the accuracy and relevance of credit assessments.

Figure 2: Epoch Model accuracy Figure 3: Epoch Model Loss

https://google.academia.edu/JournalofComputerScience 7 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

Power BI Visualisation:
Using Power BI for analysis in a credit score system (shown in figure 10) can offer valuable
insights into various aspects of credit scoring, including data visualization, trend analysis, portfolio
monitoring, and risk assessment.
Here's how Power BI can be leveraged for analysis in a credit score system:
Data Visualization: Power BI enables the creation of interactive and visually appealing dashboards
and reports that can display key metrics and trends related to credit scores. Visualizations such as
bar charts, line charts, scatter plots, and heatmaps can be used to represent credit score
distributions, credit utilization patterns, payment behaviour trends, and other relevant metrics.
Portfolio Monitoring: Power BI can be used to monitor the credit portfolio's performance in real-
time or at regular intervals. Dashboards can display metrics such as portfolio composition, credit
quality distribution, default rates, and delinquency trends. Stakeholders can track portfolio health,
identify areas of concern, and take proactive measures to manage risk.
Risk Assessment: Power BI can aid in risk assessment by integrating data from various sources,
including credit bureaus, transaction histories, and demographic information. Advanced analytics
and predictive modelling techniques can be applied within Power BI to identify high-risk accounts,
predict default probabilities, and assess the overall credit risk exposure of the portfolio.
Customer Segmentation: Power BI allows for segmentation analysis based on customer
demographics, credit profiles, and behaviour patterns. By segmenting customers into different
groups (e.g., by credit score, income level, age), financial institutions can tailor their marketing
strategies, product offerings, and risk management approaches to better serve the needs of each
segment.

IV. RESULTS
Confusion Matrix: A confusion matrix (shown in figure 4) is a helpful tool used in machine
learning to evaluate the performance of an algorithm on a classification task. The rows
represent the actual labels of the data, and the columns represent the predicted labels.

Figure 4: Confusion Matrix Figure 5: Precision- Recall Curve

https://google.academia.edu/JournalofComputerScience 8 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

Accuracy: The model shows 81% accuracy.

Precision-Recall curve: The precision-recall curve (shown in figure 5) provides a valuable tool
for evaluating the performance of multi-class classification models. It helps you assess how
well the model identifies each class and pinpoint areas for improvement.
Classification Report: A classification report (shown in figure 6) is a summary of the
performance of a classification model on a set of test data. It provides various metrics that help
you understand how well the model classified each class in your data.

Figure 6: Classification Table for Training and Validation data

Output:

Figure 7: Credit Score Predicted by the model based on Random values generated

The model has been deployed using Flask (shown in figure 8 and 9):

https://google.academia.edu/JournalofComputerScience 9 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

Figure 8: GUI for credit score prediction

Figure 9: Result

Power BI Analysis:

https://google.academia.edu/JournalofComputerScience 10 https://sites.google.com/site/ijcsis/
ISSN 1947-5500
 International Journal of Computer Science and Information Security (IJCSIS),
Vol. 22, No. 3, June 2024

Figure 10: Power BI analysis for easy understanding of the credit scoring system

V. CONCLUSION
This project has provided valuable insights into the application of neural networks, sequential
models, and SMOTE techniques in credit scoring systems. Through an in-depth investigation,
we have explored various neural network architectures and methodologies, shedding light on
their potential to enhance the accuracy and efficiency of credit risk assessment. The project
also provides visualization using Power BI for easy understanding of credit score for the
banking employees. By assessing the advantages and limitations associated with employing
neural networks in credit scoring, we have gained a nuanced understanding of their strengths
and challenges. Moreover, our examination of the challenges and opportunities inherent in
integrating these advanced techniques into existing credit risk management frameworks has
highlighted important considerations for financial institutions seeking to leverage artificial
intelligence effectively.

VI. REFERENCES
[1] A Deep Learning Approach for Credit Scoring Using Feature Embedded Transformer by
Chongren Wang and Zhuoyi Xiao (2022)
[2] Implementation of Business intelligence in Know Your Customer (KYC) for credit card
customers loan repayment status by Mario Caesar Kristantoputra and Preethi Subramanian,
(2023)
[3] A novel federated learning approach with knowledge transfer for credit scoring by Zhongyi
Wang a, Jin Xiao a, Lu Wang b, Jianrong Yao ,2023
[4] Emerging Trends in Deep Learning for Credit Scoring: A Review by Yoichi Hayashi ,2022
[5] Neural network as a tool to solve the problem of credit scoring by Vladimir K Zhirov,
Nataliya A Staroverova, Marina L Shustrova, Mariya N Tomilova ,2021
[6] Flexible loss functions for binary classification in gradient-boosted decision trees: An
application to credit scoring by Jonah Mushava, Michael Murray,2024
[7] Machine Learning Approach for Credit Score Predictions by Tsholofelo Mokheleli, Tino
Museba,2023
[8] Credit Score Prediction by Krishna Balaji, Aashima Gupta, Shristy Goswami,2023
[9] Credit-Risk Prediction Model Using Hybrid Deep—Machine-Learning Based Algorithms
by Tamiru Melese, Tesfahun Berhane, Abdu Mohammed and Assaye Walelgn,2023
[10] Analysis and visualization of financial data with power BI by Pedro Ayuso García,2023

https://google.academia.edu/JournalofComputerScience 11 https://sites.google.com/site/ijcsis/
ISSN 1947-5500