Professional Documents
Culture Documents
Loan Data Analysis Using Data Warehouse Techniques: August 2021
Loan Data Analysis Using Data Warehouse Techniques: August 2021
net/publication/353923642
CITATIONS READS
0 5,254
3 authors, including:
All content following this page was uploaded by PPG Dinesh Asanka on 16 August 2021.
Abstract— Most of the financial institutions are running deterioration and loan or credit recovery rate also reduced
their operations smoothly and profitable way without any respectively [1]. The main problem is their criteria of
interruptions with the help of data analytical techniques. accepting and rejecting borrowers with the small amount of
This study will be able to enhance the business's ability to analyzed data and mainly responsible for the bank contact.
expand its market by providing meaningful and key analysis To identify, the legitimate borrowers have to change the
of consumer behavior. Financial institutions should have screening criteria of the borrowers.
proper parameters to identify the right customer base with When the borrower applies for the loan or credit from a
the capacity of their repayments. To identify those certain finance institute whether the loan or credit is accepted
parameters, BI technologies and the data warehouse or rejected according to the screening criteria. After a certain
techniques such as inspecting, cleansing, transforming, and time period, the accepted borrower receives his/her loan. This
modeling were used to convert data to meaningful can be analyze further with the factors which we obtain by a
information. The star schema is used for this data warehouse decision tree. To accomplish the analysis by a decision tree,
design which includes one fact table surrounded by several sample data was collected by certain financial institutions. To
dimensions. This study was mainly focused to identify the discover them, Microsoft BI tools were used to do certain
borrower’s response to the calls taken by call center agents operations such as inspecting, cleansing, transforming, and
on a time basis per day. As a result, identified that several modeling to convert useful information.
parameters such as age groups and gender-wise response
times are different. Those factors will be evaluated by using A. Microsoft BI Tools.
a decision tree in future works. This will increase the loan SQL Server Integration Service (SSIS) tool which,
collection efficiency. Extract, Transform, and Load (ETL) data, which sample data
were collected from the source systems to data warehouse
Keywords—Loan Repayment, Loan Arrears, designed to analyze data. This tool helps to do certain tasks
Response Times, Decision Tree, Data Warehouse such as cleansing, filtering, conversions, sorting, joining,
validating, lookup, and aggregations. After these transform
I. INTRODUCTION operations load data to the data warehouse, which includes
Fact tables which contain measurers and Dimension tables
Borrowing credit or loans from financial institutions by which contain a set of detailed facts related to the business by
individuals, small organizations, or large organizations invest surrounding the dimensions that elaborate the business
in self-employments. Individuals who borrow a certain details.
portion of money at one time and repay that with certain
steps with the realistic short or long time period. Most of the SQL Server Analytical Service (SSAS) tool processes the
individuals borrow for education purposes or professional data. This tool is used by organizations to analyze data from
reasons with their repayment capacity calculating by multiple sources. SSAS helps to model multi-dimensional
themself with their monthly wages. Certain organizations in cubes which can measure sales amount, a quantity and so on.
the country borrow the amount from the institution to invest There are several methods such and drill down, roll up, slice,
in certain projects in which they are willing to expand their and dice to elaborate the date with the dimensions.
business or gain more profit on that. They are planning to In order to illustrate such data, this study has utilized
repay that amount with the profit and income which they will excel pivot tables and the dashboard for the managements
achieve by the project they invest over the time period. using graphs. This is a framework. Framework in the sense,
Financial institutes such as banks, leasing companies, and the user will be able to change and create their own reports or
other credit departments willing to earn their profits by illustration visualization by themselves. This is called self-
funding or giving loans or credits to such organizations or service. They will be able to change the parameters and
individuals by taking risk of repayment over the period of analyze the data that we have provided under this framework.
time they agree to. This profit-earning cycle has risk factors
and those factors should be calculate by the institution before This paper has used a decision tree to identify the
they release the amount to borrowers. categories which will affect the repayment, arrears amount,
and the response of the borrowers using the data which gives
Somehow over the repayment time period, some output from the above tools.
individuals or organizations show repayment rate
1
2021 International Conference on Advanced Research in Computing (ICARC-2021)
II. RELATED WORK 1. Business and entity risk. What is the industry of the
There are several studies conducted related to the fraud customer?
analysis of the financial institution by the borrowers or their 2. Location. Where the customer located?
customers. Loan or credit borrower’s repayment capacity 3. Business size of the customer.
and recovery may or may not lead to the fraud. 4. Products and transaction risk. Which kind of
product or services is being offered to the customer?
SU-NAN WANG and JIAN-GANG YANG have
conducted a study regarding the money laundering risk. Corresponding to the above attributes they have marked as
Criminal activities, drug trafficking, bribing, smuggling are low, middle, and high risk of the customer account
illegal activities that may lead to highly profitable areas. regarding the Anti-Money Laundering.
These will put through the cycle of a transaction before
using them freely or otherwise those will become an
illegitimate transactions. Putting through a cycle of
transactions between banks and accounts that couldn’t track
back to the illegitimate sources. This shows as legitimate
money to use them freely. That is called as money
laundering.
2
2021 International Conference on Advanced Research in Computing (ICARC-2021)
The decision tree starts with the root node which institute. The accepted loan will be received after a certain
represents the sample data set. The second step or the period of time to the applicant.
second level is called a leaf, this level illuminates measure In this paper, they have used logical regression to
and the condition of the decision tree second level. All the predict the category of the loan borrowers that find the
attributes and discrete values are classified. The attributes of best fitting model for the individual case’s relationship
the values have to be discrete. In the third step branch and between response and explanatory variables. The
the sample, values will be divide into other branches. If the Bayesian logistic regression method is used to predict the
branch has no sample, That type contains the majority of the repayment efficiency of the borrowers.
sample was created as a leaf. That defines the basic decision
tree. As a problem, if there is high repayment efficiency,
the relationship between the customer and the credit
Select the index of the research paper of this. They have institution is high due to the help of the next higher
amount obtained from the credit loan department or the
collected 21-word segments from the bank loan data set
financial institution. To change this, they considered some
from the customers. Among those, they identify 20-word
socio-economic factors. The main factors from the
segments which are relevant to evaluate the customer
institution side are tight control, loan officer intensive,
financial credit information of the customers such as check loan collection, the interest rate charged affect the
account, credit history records, savings or deposit accounts, repayment rate. The socio-economic factors from their
loan amounts, loan terms, loan purpose, controllable customer side are marital state, gender, education, and
income, sex, working time, guarantors, address, age, income level. This conducted study has answered the
property, telephone, registration, local or foreign. Finally, following questions regarding the repayment factors at the
the bank divides them into two categories as “good end of the paper [6].
customers” and “bad customers” [4].
1. What are the main socio-economic factors?
The bank thinks the salary should be the root node for the 2. What are the business and loan-related factors?
decision tree according to their viewpoint [4].
3. What are the major challenges faced by
customers and the institute?
Their data collection method was a structured
questionnaire. They have distributed the questioner among
the people with the respective population. This includes
social attributes, household characteristics, income, assets,
financial characteristics such as credit and savings.
Logistic regression analysis is an extended technique
of multiple regression analysis which is used to identify
the categorical variables. They have considered the ratio
of the probability of success, and this is the logistic model
they used,
(1)
3
2021 International Conference on Advanced Research in Computing (ICARC-2021)
The fact table consists of measures which mainly address The data warehouse was designed and referred by this
the business problem, process, and needs of the users, such as study which has one Fact table called “FactLoanCollection”
payment amounts, arrears amount, call count, inventory, with six dimension tables.
sales. The fact table contains duplicate records for the
This data warehouse consists of dimensions
analysis purpose.
DimCustomer, DimUser, DimRPCTypes, DimProduct,
A dimension table is a set of detailed business facts DimDate, and DimTime. Most of the data warehouse uses
surrounded by many dimensions that describe those facts date dimension for the performance reasons and the
such as customer attributes, location attributes, product functional reasons, that consist of the date hierarchy and
attributes, and so on. other several attributes. This study contains two hierarchies
for that with quarter and month name as well as numbers for
B. Data Set the user’s perspective. Apart from the date dimension, time
The sample data set was collected for the analysis from dimension was added, which helps to analyze a data on an
the financial institution of a particular country. Among the hourly basis, due to the sample data set that identified it is
larger scale of data set factors were identified to analyze, importance to analyze the data of the user’s response time per
such as profession, sex, age range, locations wise. The the calls. This particular time dimension consists of the
borrower’s repayment amounts, arrears amount, arrears days hierarchy call HourMinSec that contains hour, minute, and
for the particular month or quarter, response per call, and seconds. RPCType dimension contains two attributes that
promise to pay amount particular customer range were response per call type code and the description of that code.
derived from that source of data set. The product dimension consists of the borrower’s loan types.
Before loading data to the data warehouse, there was a Furthermore, In Respect to the SCDs, This study has used
workaround to accomplish the task such as cleansing data, Type 1 SCDs most of the dimensions
conversion, filtering as per needs, sorting, joining,
aggregating, and lookup.
4
2021 International Conference on Advanced Research in Computing (ICARC-2021)
Similarly, this framework also consists of OLTP data perform truncation prior to load fact. Also, the cube
source that loads data to the warehouse which is designed via dimension and the fact process are also done by this stage.
ETL technology and process cube to browse dAata as This will be a very convenient process if the developer runs
multidimensional with several technologies that enable on the schedule of those tasks. No one needs to run every
slicing, dicing, roll-up, drill-down, and pivot. task manually. The administration just has to publish the
ETL package to the package store and enable the SQL job for
us. That process will run every time that you have scheduled
previously. Apart from that, it will be able to add the task to
send a notification to the administration or users.
E. Analysis
Data Analysis is a process to discover useful information
for business decision making. This is used to extract useful
information from the raw data to make a day-to-day decision
based on that information. That may vary according to
business needs or individual needs. Analysis is most of the
time based on the future prediction by looking into past
Figure 7. Architectural design diagram of Loan data analysis information or patterns that we have discovered, something
like forecasting whether and the environmental factors.
D. ETL (Extract, Transform, Load)
There are several types of data analysis techniques such
as Text analysis, statistical analysis, predictive analysis,
ETL is a process of load data from source to destination. diagnostic analysis, and prescriptive analysis.
ETL defines Extract, Transform, and Load. Extract in the
sense extracting data from the source. That source may not There are several tools to accomplish the data analysis
be an SQL environment that could be several types of task, this study has used Microsoft SQL Server Analysis
sources. In this operation, we will face extracting data from Services (SSAS) [11] to deliver our framework for the users.
many systems platforms such as flat files, web servers,
emails, images. Due to this we are extracting several types of F. Reporting
data into one database called staging area. You have to do Reporting means illustrating information in a proper
cleaning and changes to the operational data that simplifies understandable manner to users. There are several technical
the building summaries that help to avoid slowness, security tools such as excel, PowerBI reporting, SQL Server
matters, impossibility, also it avoids the conflicts between Reporting Service [9], Tableau so on. In this approach, we
source systems. That can be used as a backup option as well. have used excel features to represent the dashboard with the
The second stage of the ETL is transform, it consists of pivot charts.
several rules before load data to the destination. With this This study was conducted to identify the best time slots to
stage filtering, sorting, pivoting, validating, deriving make an effective customer call by a call center agent for
columns, joining or merging, splitting, lookups operations financial institutes other than collecting a lot of information
can be performed. The above rules vary between the business about loan borrowers.
need. The approach mostly uses derive functions in this stage
to derive and convert data to certain types.
Third and final stage is loading. This stage will check the
error logs send a notification upon that to the administration
or the users regarding the availability of the data that can be
performed after loading data into the destination.
This study has chosen Microsoft BI SQL Server
Integration Service (SSIS) tool to perform the ETL process.
5
2021 International Conference on Advanced Research in Computing (ICARC-2021)
TABLE I : Response Per Call Age Group Wise. past data by analyzing them. This study was implemented in
a framework. It enables to fetch the reports as per the user
Age requirement with the available data set. For this, Microsoft
Hour Groups BI tools were used in order to develop ETL and Cube. Excel
20-29 30-39 40-49 50-59 > 60 enables access of this framework for the users, mainly for
the management to make their decision, on that they will be
7 0.10% 0.15% 0.13% 0.29% 0.00% able to reduce their risks by giving loans or credits to certain
8 4.14% 4.03% 3.28% 4.49% 4.43% borrowers. This study was conducted to identify the
9 10.26% 10.45% 10.07% 10.49% 9.03% borrower’s response as per the call on a time basis per day.
10 10.63% 11.06% 11.17% 11.37% 11.12% As a result, the researcher was able to identified different
key response times of loan borrower’s based on several
11 11.25% 11.55% 11.97% 12.03% 13.04% parameters such as age groups and gender. This study varies
12 5.87% 5.97% 6.17% 6.04% 7.86% from the other studies by recognizing the best time slots to
13 11.62% 11.06% 11.99% 11.21% 11.45% make an effective customer call by a call center agent for
14 11.72% 11.35% 11.82% 10.40% 11.71% financial institutes. For future work, we have decided to
implement a decision tree-based algorithm with
15 11.12% 10.81% 10.37% 10.91% 10.20% classification and regression techniques.
16 9.23% 9.73% 10.15% 10.05% 8.36%
REFERENCES
17 6.66% 6.86% 6.46% 6.32% 5.94%
18 5.05% 4.82% 4.57% 4.56% 4.52% [1] AbrehamGebeyehu (2002). “Loan repayment and its
19 2.36% 2.15% 1.83% 1.85% 2.34% Determinants in Small-Scale Enterprises Financing in
Ethiopia: A Case of Private Borrowers AroundZeway
G. Non Functional Requirements Area”, M. Sc. Thesis, Addis Abeba University.
Very important factors include the data analysis, that [2] Ted E. Senator, Henry G. Goldberg, Jerry Wooton, etc.,
The financial crimes enforcement network AI system
contains the non-functional requirements, which most of the (FAIS) identifying potential money laundering from
parties do not consider and may be not known. This can be reports of large cash transactions[J], AI Magazine,
defined as quality attributes, such as usability, availability, Vol.16, No.4, pp. 21-39, Winter 1995.
security, and reliability. Those important factors must be [3] Safavin,S.R., Landgrebe,D. A survey of decision tree
involved with the data. If we are working with banking data classifier methodology [J]. IEEE Transactions on
sources those should be considered mainly due to security. Systems, Man and Cybernetics, Vol.21, No.3, pp.660-
667, April 1991
Availability which means that data should available any time
[4] SU-NAN WANG, JIAN-GANG YANG. A MONEY
anywhere to access for the users with the credentials they LAUNDERING RISK EVALUATION METHOD
provide, credential? Yes, when we get with the credentials BASED ON DECISION TREE, College of Computer
first thing that comes up is security. The majority nowadays Science and Engineering, Zhejiang University,
most organizations consider those factors. Analysis systems Hangzhou 310027, China 2Shanghai Pudong
with nonfunctional requirements provide a better framework. Development Bank, Shanghai 200002, China 2007.
[5] JI Chengjun, WU Lijun, LI Jinping. The Application of
the Decision Tree Analyzes in the Credit Card,
IV. FUTURE WORK Department of Management, Liaoning Technical
University, Huludao, China
The financial institutions will reduce their risk factor if
[6] Yonas Shuke Kitawa, Nigatu Degu Terye.
they have a proper evaluation of the loan or credit borrowers. (2020/10/30). Statistical Analysis on the Loan
Developing a proper classification algorithm with a decision Repayment Efficiency and it’s impact on the
tree will be able to avoid such a risk. To identify, the Borrowers: A case study of Hawassa city, Ethiopia
borrowers can develop a popular data mining technique. Available
Decision tree is based on [8] algorithm with the factors such :http://article.sciencepublishinggroup.com/html/10.1164
as gender, profession, and location of the borrowers 8.j.ajtas.20150406.28.html
according to the training data set. [7] (202/09/13), Decision tree algorithm example in data
mining. Available :
A decision tree is a flowchart, it is like a tree. This is a https://www.softwaretestinghelp.com/decision-tree-
top-down approach. This structure is made on training data algorithm-examples-data-mining/
set, those data are broken into a small set of subsets and [8] Wang, J. Data Warehousing and Mining: Concepts,
illustrate in the nodes of that tree. The decision tree has a root Methodologies, Tools, and Applications. Hershey:
node, branches, leaf nodes. The decision tree uses Information Science Reference, 2008.
classification and regression which are used to predict and [9] Microsoft (2020/07/22). Analysis and reporting with
divide the nodes. The classification has been used for Microsoft business intelligence (BI) tools. Available :
https://docs.microsoft.com/en-us/sql/reporting-
categorizing data set into certain parameters [7]. Regression services/choosing-microsoft-business-intelligence-bi-
is used for the prediction of the values, this is used to predict tools-for-analysis-and-reporting?view=sql-server-ver15
the model’s continuous values. [10] Create an Extended Date Dimension for a SQL Server
Data Warehouse, Available :
https://www.mssqltips.com/sqlservertip/5553/create-an-
V. CONCLUSION extended-date-dimension-for-a-sql-server-data-
In conclusion, some financial institutions are facing warehouse/. Dinesh Asanka, MSSQLTips.com.
a huge risk due to the failure of recovering loans and [11] What is Analysis Services?,
https://docs.microsoft.com/en-us/analysis-
collections on time. To avoid this and make decisions on the services/analysis-services-overview, Microsoft.
6
View publication stats