157
Reference:
1.
Mirziyoev Sh., appeal of the president of the Republic of Uzbekistan Shavkat Mirziyoev to
the Assembly Oliy Majlis, People's word, 29.12.2020.
2.
Decree of the president of the Republic of Uzbekistan dated 28.01.2022 "on the
development strategy of new Uzbekistan for 2022
–
2026" No. 60.
3.
Nguyen, Khoa Huu, A. Coronavirus Outbreak and Sector Stock Returns: The Tale from The
First
Ten
Weeks
of
2020
(March
25,
2020).
Available
at
SSRN:
https://ssrn.com/abstract=3559907
or
http://dx.doi.
org/10.2139/ssrn.3559907:
https://insight.factset.com/sp-500-earnings-season-update-april-17-2020.
APPLICATION OF MACHINE LEARNING IN DETECTING LOAN DELINQUENCY:
CASE STUDY OF MICROFINANCE INSTITUTION IN UZBEKISTAN
Isakov Olmas Kuchkarovich
PhD candidate in Econometrics and Statistics
Westminster International University in Tashkent
The rise of the internet has revolutionized the way we live, work, and
communicate. Alongside this digital revolution, a new phenomenon has emerged
- big data. Big data refers to the vast amount of structured and unstructured
information generated by individuals, organizations, and devices. Big data's
emergence has brought about transformative changes across various industries.
In particular, big data analytics has enhanced risk management, fraud detection,
and personalized banking experiences for customers in the finance industry. The
application of the old credit scoring model has been severely constrained by the
growth of big data on the Internet, and the original business logic framework has
been lost under the new data profiles and business situations.
The modern banks must implement a variety of machine learning (ML)
techniques to reduce the manual involvement in the monitoring and testing
process and utilize improved automated approaches to deal with increasing
number of potential borrowers. Subjective judgement of credit experts on loan
advancements is very inefficient and can be dependent on the decision-making
ability of those experts. Therefore, application of statistical and machine learning
methods can address these problems and might be important tools in credit risk
management for several reasons. First, machine learning algorithms have the
ability to analyze vast amounts of data and identify patterns that may not be
apparent to human analysts.
They can even identify complex relationships and non-linear patterns in data
that may not be detectable by human experts. This can lead to more accurate
predictions of creditworthiness and a better understanding of the potential risks
associated with a particular borrower. Commercial banks can tailor credit offers
to individual borrowers' risk profiles, leading to more accurate pricing and better
risk management. Moreover, machine learning algorithms can quickly analyze
and process large volumes of data in real-time. This enables lenders to make
158
faster and more informed decisions about credit applications, reducing the time
and effort required for manual analysis. By automating the credit risk assessment
process, machine learning can reduce the need for manual analysis, saving time
and resources for lenders. This allows them to process a larger volume of credit
applications and improve operational efficiency.
Recent research has examined the use of several machine learning
techniques, including decision trees, neural networks, support vector machines,
and integration algorithms, in the evaluation of credit. [1] reported that classical
logistic regression did not perform as well as the other machine learning methods
when there are nonlinear relationships between the variables. However, the
logistic regression model offers a significant benefit in terms of variable
interpretability and stability, even though prediction accuracy may not be as good
as that of the machine learning model.
As a result, some researchers enhanced the logistic regression and used it to
forecast borrower default behavior. In [2] personal credit risk assessment was
conducted using five famous machine learning methods such as N
aïve Bayesian
Model, logistic regression model, random forest decision tree and K-Nearest
neighbor classifiers. A personal credit evaluation model based on the naive
Bayesian classifier was first proposed by [3]. It was tested on German and
Australian credit data sets and compared with five neural network models,
showing that the naive Bayes classifier has a lower classification error, hence
higher accuracy rate. [4] evaluated the P2P online loan borrowers using a neural
network model. The results demonstrated that the model was capable of good
feature extraction and knowledge discovery.
The borrower evaluation index system can still produce a more accurate
assessment of the credit risk of borrowers and has a good capacity to assess and
anticipate when virtual information index is present. [5] have used a real social
lending platform (Lending Club) dataset with more than 800,000 observations
considering different evaluation metrics (i.e. AUC, Sensitivity, Specificity) using
random forest, logistic regression and multilayer perceptron. Besides supervised
learning techniques, several researches have focused on clustering-launched
support vector machine (SVM) models using unsupervised ML algorithms like
divisive hierarchical k-means (DHK) and self-organizing maps (SOM) ([6] and
[7]).
In this paper, the dataset has been obtained from a micro finance institution
(MFI) in Uzbekistan which is well established and has branches in all 14 regions
of the country. The dataset contains 12883 individuals (clients) for the period
2019-2022. The summary statistics of the observations are provided in Table 1.
The following statistical method was applied to estimate the probability of
being delinquent on the loan payments.
ln(
𝜋
1−𝑝
) = 𝛽
0
+ 𝛽
1
∗ 𝐺𝑒𝑛𝑑𝑒𝑟 + 𝛽
2
∗ 𝐴𝑔𝑒 + 𝛽
3
∗ 𝑁𝑢𝑚
𝑙𝑜𝑎𝑛𝑠
+ 𝛽
4
∗
𝐴𝑚𝑜𝑢𝑛𝑡 + 𝛽
5
∗ 𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑟𝑎𝑡𝑒 + 𝛽
6
∗ 𝐴𝑔𝑒 𝑠𝑞𝑢𝑎𝑟𝑒𝑑 + 𝜀
159
Table 1.
Descriptive statistics of the dataset obtained from MFI
The data was split into train (70% of the dataset) and test (30% of the
dataset) in order to compute the performance of the model. The train data was
used to build the model using the variables provided in Table 1 and the test data
was used to predict the probability of being delinquent. Based on the predictions
and classification of borrowers into delinquent (when probability of delinquency
is higher than 0.5) and not delinquent, the proposed model produced 88%
accuracy, 6% sensitivity and 99% specificity. The results of the logistic regression
model are provided in Table 2.
Table 2.
The results of logistic regression model
* - significant at 10%, ** - significant at 5%, *** - significant at 1%.
According to the results, the male borrowers are more likely to be delinquent
on the loan payments while holding other variables constant. Age variable and its
squared transformation did not show any significant impact on the likelihood of
delinquency. Number of loans, amount of the loan and interest rate showed
positive relationship with the probability of being delinquent. These findings can
have significant implications for policymakers as well as financial professionals
in Uzbekistan in tackling issues related to non-performing loans.
References:
1.
Galindo, J., & Tamayo, P., 2000. Credit risk assessment using statistical and machine
learning: Basic methodology and risk modeling applications. Computational Economics 15, 107
–
143.
2.
Wang, Y., Zhang, Y., Lu, Y., & Yu, X., 2020. A Comparative Assessment of Credit Risk Model
Based on Machine Learning a case study of bank loan data. Procedia Computer Science, 174, 141
–
149.
160
3.
Xusheng L., & Yaohuang G., 2006. Personal credit evaluation model based on Naive Bayes
classifier [J]. Computer Engineering and Applications, 42(30): 197-201.
4.
Kim S.H., Oh K.J., Ju J.B., & Lee D.W., 2019. Predicting Debt Default of P2P Loan Borrowers
Using Self-Organizing Map. Quantitative Bio-Science, 38(1), 63
–
71.
5.
Moscato, V., Picariello,
A., & Sperlí, G., 2021. A benchmark of machine learning approaches
for credit score prediction. Expert Systems With Applications, 165, 113986.
6.
Luo, S., Cheng, B., & Hsieh, C., 2009. Prediction model building with clustering-launched
classification and support vector machines in credit scoring. Expert Systems with Applications, 36
(4), 7562
–
7566.
7.
Yu, L., Yue, W., Wang, S., & Lai, K. K., 2010. Support vector machine based multi- agent
ensemble learning for credit risk evaluation. Expert Systems with Applications, 37 (2), 1351
–
1360
.
EXPLORING THE CREDIT TRABSFER SYSTEM IN TEACHING
ENGLISH FOR ECONOMICS
Kurbonova Nigora Ne’matovna
ESP teacher “English Language” department TSUE
The credit module system is a method of teaching languages that is gaining
popularity around the world. This system is designed to provide students with a
more flexible and personalized approach to language learning. In this article, I will
explore the prospects of the credit module system of language teaching, including
its benefits and challenges. Firstly, let us understand what the credit module
system is and its history background. The credit module system, also known as
the credit-based system, is a method of measuring academic achievement based
on the number of credits a student earns for completing a particular course or
program. The system has been in use for many decades, but its origins can be
traced back to the United States in the late 19th century.
The credit hour was initially introduced in 1906 at the University of Chicago
by the university's president, William Rainey Harper. He believed that the credit
hour would provide a flexible means of measuring academic achievement,
enabling students to take courses in a variety of subjects and earn credit towards
their degree. The credit hour system quickly became popular among other
universities in the United States and was adopted by the Carnegie Foundation in
1910 as a standard unit of academic measurement [3]. Over the years, the credit
module system has been refined and adapted to meet the changing needs of
higher education. In the 1960s, for example, the system was modified to include
the concept of modularization, which allowed students to take individual modules
of courses rather than having to complete an entire course to earn credit.
This made it easier for students to design their own courses of study and to
tailor their academic programs to their specific needs and interests. In the 1970s
and 1980s, the credit module system was further developed to include the idea of
credit transfer. This meant that students could transfer credits earned at one
