ANSWER SCRIPT

FINAL ASSESSMENT MAY TRIMESTER

2023/2024
PROGRAMME :
MASTER OF BUSINESS ADMINISTRATION
MATRIC NO. : PBS23101013
COURSE : MGT7105 BUSINESS
INTELLIGENCE AND
ANALYTICS SYSTEMS
DATE OF FINAL ASSESSMENT : 8 AUGUST 2024 – 12.00
PM (MYT)
SUBMISSION DEADLINE : 10 AUGUST 2024 – 12.00 PM
(MYT)
TIME ALLOCATION : 48 HOURS (2 DAYS)

INSTRUCTIONS:
1. Please use the answer script template to answer the questions
or foolscap paper if the answer requires calculation or
drawing.
2. Please write your Matric No., Code and Course Name in the
answer script or Foolscap papers. The foolscap papers
containing the answers need to be scanned/captured and
attached to the email.
3. All answer scripts must be emailed to
[email protected] within the scheduled time
given.
4. Before emailing your answer scripts to the Examination Unit,
please SAVE the answer scripts with details of your Matric
No. & Course Code as shown below:
Example:
PBS2020XXXX_HPM7205
5. The answer scripts must be saved in WORD format.
6. Please indicate your Course Name and Matric No. as the
subject heading when emailing your answer script.
7. The similarity percentage must be ≤30% as shown below:
 No. Similarity Percentage Marks Deduction
(%) (%)
1 0-30 0
2 30.01 – 100 FAILED
Note: Kindly take note that plagiarism is a form of cheating that
will not be tolerated. This includes using Artificial Intelligence
(AI) and external sources such as ChatGPT, any translation or
paraphrasing tools, internet browsing, downloading, and
referencing notes. A student who is caught cheating will be
automatically given an F grade and will have to retake the
course.
*Disclaimer: The similarity index for plagiarism is subject to
the course offered.
8. Cheating or attempts at cheating violate the
University’s regulations.
-Good Luck-
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

Section A

Proposal: Utilizing AI in Business Analytics to Detect Fraudulent Activities in

Banking

1. Purpose of this Proposal

The purpose of this proposal is to provide banks with a detailed plan for using
artificial intelligence (AI) to identify possibly fraudulent activity. The proposal will go
over the strategies to be used, their significance in the banking industry, and how to
apply them successfully to protect financial activities.

2. Strategies of Fraudulent Activities Detection

2.1 Descriptive Analytics

The foundation of fraud detection is descriptive analytics, which

focuses on examining past data to find patterns and trends that can point to
fraud. A thorough understanding of the current situation of customer
relationships is offered by descriptive analytics, which looks at customer
behaviours, transaction histories, and interactions with banking services.

2.2 Predictive Analytics

Descriptive analytics lays the groundwork by analysing historical data

to identify patterns and anomalies in customer behaviour. Using data mining
techniques, predictive analytics looks for hidden trends that may point to
fraud. They assist in finding connections that may not be immediately
apparent, such as a combination of specific transaction kinds, frequencies, or
locations that usually occur before fraudulent behaviour. Once these patterns
are found, they can be utilized to create predictive models that, when updated
with new data, can forecast the probability of fraudulent transactions. These
models do not predict a specific fraudulent event before it happens but rather
assess the likelihood that a particular transaction or set of behaviours is
fraudulent based on past data.

2.3 Prescriptive Analytics

Prescriptive analytics makes recommendations for the best course of

action based on the results of predictive model analysis. This strategy is
essential for guaranteeing that the bank responds to possible fraud in a timely
and appropriate manner.
1

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

3. Strategies Implementation

3.1 Implementing Descriptive Analytics

To identify patterns that may be signs of fraud, banks must compile a

large amount of data from both internal and external sources. The data is
primarily gathered internally from customers via a variety of methods. This
comprises transaction data which is essential for building a framework for
identifying fraud. Banks should also retrieve data such as:

● Demographic information
● Account types
● Login times and locations
● Service usage patterns
● Payment methods

This information may be acquired by banks through call centers, ATM

withdrawals, etc. Digital banks could use channels like mobile apps, and
online banking portals to obtain comparable data. For a more comprehensive
context, sources such as industry-wide fraud databases or credit reporting
organizations can also be integrated. The goal is to compile a comprehensive
dataset that reflects all aspects of customer interactions.
Every piece of information gathered needs to be integrated into the
data warehouse so that descriptive analytics methods can be applied to it.
Key patterns and outliers can be highlighted in a dashboard automatically
created with generative AI. Key metrics are represented visually by charts,
graphs, tables, and other features seen in the dashboard. The goal is to give
users a straightforward overview so they can quickly understand the state of
transactions.

3.2 Implementing Predictive Analytics

3.2.1 Discover Hidden Patterns

Data mining examines particular data points e.g. historical

transactions) to find patterns that could indicate fraud. From these particular
cases, the AI model generalizes to build prediction models that evaluate the
probability of fraudulent activity in the future. This process is called inductive
reasoning.

a. Clustering

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

Using shared attributes, similar transactions or consumer

behaviours can be grouped into clusters. Clustering could show a
pattern of consistently low-value but often occurring transactions; this
may point to a money-laundering tactic used to avoid fraud detection.

b. Classifying

Classification entails grouping transactions into pre-established

classifications (such as "fraudulent" vs. "non-fraudulent"). Banks can
identify the traits of fraudulent transactions and use this information to
categorize new transactions by training a classification model on
transaction data from the past. For example, a classification model
could be trained to identify that there is a higher chance of fraud
associated with significant transactions launched from unfamiliar
locations or new devices.

c. Association Rule Learning

This method reveals correlations or links between several data

variables. For instance, it can show that a certain pattern of behaviour
—like several login attempts followed by a sizable withdrawal—is
frequently connected to fraud. Transactions that match these patterns
can subsequently be flagged using these association rules.

3.2.2 Utilize Hidden Patterns

a. Risk Scoring

Based on the probability that a transaction will resemble known

fraudulent patterns, predictive models can assign a fraud risk score to
each transaction. For example, if a transaction corresponds with a
high-risk cluster (such as unusual transaction amounts from a
particular region), the model assigns a higher fraud risk score to it.
Transactions with scores above a predetermined threshold may be
automatically flagged for further investigation.

b. Customer Behaviour Analysis

Predictive models identify anomalies from regular behaviour

patterns that might point to fraud by continuously evaluating customer
behaviour. The predictive model would identify a possible fraud risk, for
example, if a customer who typically performs modest, local
transactions suddenly makes a significant international transfer.
3

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

3.3 Implementing Prescriptive Analytics

3.3.1 Decision-Making Process

Analysing data from predictive models that have recognized

possible fraud threats is the first step in prescriptive analytics. It
assesses the context to decide on the appropriate course of action, for
example, if predictive analytics flags a transaction as suspicious
because of its size, origin from a new device, or unusual timing.
The system evaluates many variables, such as the transaction's
possible impact, the customer's past transaction history, overall risk
profile, etc. This evaluation aids in figuring out how serious the risk is
and how much intervention is required.

3.3.2 Recommending Actions

Prescriptive analytics assists banks in streamlining their

response plans by assessing various course of action options and
identifying the most efficient way to minimize fraud without jeopardizing
the interests of real customers. This could entail choosing whether to
automatically prohibit a transaction versus when to need further
verification to strike a balance between security requirements and
customer convenience.

3.3.3 Implementation through Rule-Based Systems

A rule-based system is used to put the recommendation from

prescriptive analytics into practice. These predetermined guidelines
comply with legal requirements, fraud detection trends, and industry
best practices. For instance, a rule might stipulate that the customer
must confirm each transaction from a new device that exceeds a
specific amount before it can be processed.
If a large transaction from a new device is flagged as suspicious,
AI might recommend holding the transaction and contacting the
customer for confirmation. After that, the rule-based system carries out
the suggested course of action, making sure that any possible fraud is
dealt with before it is conducted. The AI uses deductive reasoning to
decide on particular transactions by applying the general rule (if X, then
Y).

3.3.4 Continuous Learning and Optimization

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

Prescriptive analytics systems improve their decision-making

processes over time by learning from the results of their suggestions.
The algorithm gives precedence to comparable actions in the future if
they show to be especially successful in stopping fraud.
Real-time monitoring of all transactions and activities is done
continually by the AI model. To do this, every transaction must be
examined in real-time and evaluated in light of standards, risk factors,
and patterns. The AI algorithm makes sure it can quickly identify even
minute indications of fraudulent behavior by scrutinizing every activity.
AI models can learn from any errors they make, including false
positives, which identify genuine transactions as fraudulent, and false
negatives, which fail to identify real fraud. The AI model lowers the
likelihood of future failures by refining its algorithms through the
analysis of existing errors.
The AI model can change its rules in response to the constant
stream of new data. This ongoing adaptation allows the model to check
every new activity with updated knowledge. This guarantees that the
approach will continue to work even if fraudsters change their
strategies.

Number of words: 1314

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

Section B

a) Approaches in Building Data Warehouse

There are two approaches to the building of a data warehouse. The first is the
Inmon Model: EDW approach, whereas the second is the Kimball Model: Data Mart
approach. This is an overview of how these approaches compare when developing a
data warehouse.

1. Inmon Model: EDW Approach

The Enterprise Data Warehouse (EDW) model is ideally suited for
large enterprises with long-term strategic objectives for their data warehouse
and a need for comprehensive, organization-wide data integration. The EDW
strategy seeks to provide a consolidated and unified data repository that
facilitates cross-functional decision-making by including a wide range of topic
areas throughout the entire organization.
With its emphasis on enterprise-wide coverage, the EDW strategy is by
its very nature very complicated. Large volumes of data, from gigabytes to
petabytes, are managed by it, necessitating a great deal of effort in the areas
of architectural development, data standards, and integration. Since the
model combines information from many internal and external systems,
common data standards must be followed across the entire business to
guarantee consistency and smooth integration.
In addition to requiring an extensive amount of time, the construction of
an EDW requires a significant financial and resource commitment. Despite
these challenges, the EDW model is the best option for businesses looking to
make strategic, data-driven decisions from an all-encompassing, enterprise-
wide viewpoint. It also provides a scalable solution that can adapt to the
company's changing needs and future expansion.

2. Kimball Model: Data Mart Approach

The Data Mart Approach is especially well-suited for businesses
seeking a more flexible and business-focused approach to data warehousing.
This strategy places a strong emphasis on creating data marts that are
customized to satisfy the unique requirements of different departments or
business divisions, including marketing, finance, or sales. Every data mart
offers quick and focused insights, maximizing decision-making within its own
subject area.
When compared to the EDW model, the Data Mart Approach's
implementation speed and relative simplicity are two of its main advantages.
The data mart strategy is less complicated because it concentrates on
particular subject areas and works with smaller datasets, which usually range
6

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

in size from megabytes to several gigabytes. This enables development to

proceed more quickly—often in a matter of months—and with less initial
expenditure. The methodology works especially well for businesses that don't
have the luxury of time to integrate all of their data upfront and need to see
quick wins and improvements in particular business areas.
Because the Kimball Models are developed gradually, the methodology
is less resource-intensive and more flexible. Common data standards are
maintained inside each data mart's specialized business sector, but they are
not always required to be standardized throughout the entire organization
because each data mart incorporates data from relevant operational systems.
This makes it possible to adapt and customize to each department's particular
needs.
The Data Mart Approach is a great option for businesses looking for
immediate results and localized optimization, even though it might not provide
the same degree of enterprise-wide integration as the EDW approach. It
offers an affordable, scalable solution that can develop with the addition of
more data marts over time to create a more complete data warehouse.
Businesses with limited resources or those who need quick deployment on
decisions made within certain business operations can especially benefit from
this approach.

Table 1: EDW Approach and Data Mart Approach Comparison

Inmon Model: EDW Kimball Model: Data Mart
Criteria Approach Approach
Ideal For Large companies Smaller companies
Scope Enterprise-wide, covering a Department-specific, focusing on
wide range of subject areas. the needs of individual business
areas.
Complexity High complexity due to large Lower complexity; focuses on
volumes of data, architectural specific subject areas with smaller
development, and integration. datasets.
Data Volume Manages large-scale data Works with smaller datasets
Cost Significant financial and Lower initial cost, more budget-
resource commitment. friendly.
Integration Integrates data from many Integrates data from relevant
internal and external systems operational systems within specific
across the enterprise. business areas.
Scalability Scalable for future growth and Scalable through incremental

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

changing needs of the addition of data marts over time.

company.
Best Suited For Strategic, data-driven Immediate results and localized
decisions from an enterprise- optimization within specific
wide viewpoint. departments.
Implementation Resource-intensive, requiring Less resource-intensive, more
extensive time and effort. flexible, and quicker to implement.

b) Dimensional Modelling

When designing a data warehouse, choosing the right dimensional modelling

approach is crucial to ensuring efficient data retrieval, easy maintenance, and
optimized performance. The Star Schema and Snowflake Schema are two of the
most widely used dimensional models. Below, each schema, highlighting its
features, criteria for selection, and appropriate use cases are discussed.

Image 1: Star Schema (Left) and Snowflake Schema (Right)

1. Star Schema

When the data model is simple and has few hierarchical interactions
between dimensions, Star Schema should be used (see image 1). The central
table contains metrics that the business wants to analyse. Multiple dimension
tables that surround this fact table give the data in the table a descriptive
context. Dimension tables are usually denormalized, which means that to
minimize the number of joins required in queries, they contain redundant data.
Denormalized schemas require fewer joins during querying, which leads to
faster query performance—particularly for large datasets.

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

Star Schema is a recommended schema for most data warehouse

deployments because of its optimal query efficiency, ease of use, and
simplicity. It works especially well if the main objective is to give business
users quick, easy access to data via dashboards and reports. Star Schemas
are a popular option for front-end analytics because of their easy-to-
understand and intuitive structure, which even non-technical users can use.

2. Snowflake Schema

The Snowflake Schema should be utilized when the data model has
intricate linkages and hierarchies between the dimensions (see image 1). Like
the Star Schema, the central table has the metrics that the company wants to
examine. The surrounding dimension tables in a Snowflake Schema, in
contrast to a Star Schema, are normalized into several related tables. By
dividing dimension tables into smaller, more focused tables that reflect
hierarchical relationships, this normalizing technique lowers data redundancy.
When executing queries in a Snowflake Schema, the normalized
structure necessitates additional joins, which may lead to marginally slower
performance than in a Star Schema. On the other hand, because less
redundant data means less total storage needed, it has a major positive
impact on storage efficiency. This is because common attributes (such as
category names) are saved only once in their respective tables after the data
has been normalized, as opposed to being repeated over several rows in a
single, denormalized table. This implies that the category name is only saved
once and accessed by a key, as opposed to being stored for each product.
When working with intricate data models that have intricate hierarchies
and when storage efficiency is a top concern, Snowflake Schema is advised.
Large-scale data warehouses are a good fit for it because of the complexity of
the data, which calls for a more organized and standardized approach. Due to
its higher number of joins, the Snowflake Schema may necessitate more
sophisticated query authoring and maintenance; yet, it offers a versatile and
space-efficient solution that can successfully handle intricate data exploration
and complicated analytical queries.

In summary, the Star Schema is the recommended option for the majority of
data warehouse implementations that place a high value on query efficiency and
user-friendliness. However, the Snowflake Schema ought to be used when working
with intricate data structures and there's a pressing requirement to maximize storage
efficiency. The choice of these schemas should take into account the unique needs
of the organization, striking a balance between data complexity management,
performance, and storage efficiency.

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.
 FINAL ASSESSMENT MAY TRIMESTER
2023/2024
MATRIC NO.: PBS23101013
COURSE :
MGT7105 BUSINESS INTELLIGENCE AND ANALYTICS SYS

10

*Note: Use only the same typeface and font size as below:
Typeface: Arial
Font size: 12-point.