CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

CMA INTERMEDIATE | GROUP 2 | PAPER 11

Financial Management & Business Data Analytics
SECTION B – Business Data Analytics | 20% Weightage

MODULE 11
Data Analysis and Modelling
Sections: 11.1 Process, Benefits & Types | 11.2 Data Mining | 11.3 Analytics & Model
Building (Descriptive, Diagnostic, Predictive, Prescriptive) | 11.4 XML & XBRL | 11.5
Cloud, BI, AI, RPA, ML | 11.6 Model vs Data-driven Decision Making

MODULE LEARNING OBJECTIVES

⬤ Understand the process, development and types of data analysis.

⬤ Understand the concepts of data mining and implementation of data analysis.

⬤ Understand the concepts of analytics and model building.

⬤ Understand the standards for data tagging and reporting.

⬤ Understand the concepts of cloud computing, business intelligence, artificial

intelligence, robotic process automation and machine learning.

⬤ Understand the relationship between model and data driven decision making.

11.1 Process, Benefits and Types of Data Analysis

WHAT IS DATA ANALYTICS?

Data analytics is the science of evaluating unprocessed datasets to draw conclusions about
the information they contain. It helps identify patterns in the raw data and extract useful
information. Applications containing machine learning algorithms, simulation and automated
systems are utilised. These data are evaluated and used to assist firms in gaining a deeper
understanding of their customers, analysing promotional activities, customising content,
developing content strategies and creating new products. Data analytics enables businesses
to boost market efficiency and increase profits.

11.1.1 PROCESS OF DATA ANALYTICS — 5 STEPS

Step Stage Description

Step Criteria for Data may be segmented by a variety of parameters, including
1 Grouping Data age, population, income and sex. Data values might be either
numeric or category.
Step Collecting the Data may be gathered from several sources, including internet
2 Data sources, computers, personnel and community sources.

The Institute of Cost Accountants of India Page 1

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

Step Stage Description

Step Organizing the After collecting the data, it must be arranged so that it can be
3 Data analysed. Statistical data can be organised on a spreadsheet or
other programme capable of handling statistical data.
Step Cleaning the Data The data is initially cleansed to verify that there are no
4 duplicates or errors. The document is then examined to ensure it
is comprehensive. Before data is sent to a data analyst for
analysis, it is beneficial to rectify or eliminate any errors.
Step Adopt the Right Choose from four types of data analytics process: (i) Descriptive
5 Analytics Type analytics (ii) Diagnostic analytics (iii) Predictive analytics (iv)
Prescriptive analytics.

11.1.2 BENEFITS OF DATA ANALYTICS

Benefit Description
(i) Improves Companies can use information from data analytics to base their
Decision Making decisions, resulting in enhanced outcomes. Reduces guesswork
involved in preparing marketing plans and deciding what materials to
produce. Continuously collect and analyse new data to gain a deeper
understanding of changing circumstances.
(ii) Increase in Data analytics assists firms in streamlining their processes,
Operational conserving resources, and increasing profitability. When firms have a
Efficiency better understanding of their audience's demands, they spend less
time creating advertising that does not fulfil those needs.
(iii) Improved Data analytics gives organisations a more in-depth understanding of
Service to their customers, employees and other stakeholders. This enables the
Stakeholders company to tailor stakeholders' experiences to their needs, provide
more personalisation and build stronger relationships.

11.2 Data Mining and Implementation of Data Mining

WHAT IS DATA MINING?

Data mining, also known as Knowledge Discovery in Data (KDD), is the extraction of
patterns and other useful information from massive data sets. Given the advancement of data
warehousing technologies and the expansion of big data, data mining techniques have
advanced dramatically over the past two decades, supporting businesses in translating their
raw data into meaningful information. Data mining has enhanced corporate decision making
through smart data analytics. The two primary purposes: (1) Describing the target dataset;
(2) Predicting results using machine learning algorithms. These strategies surface the most
relevant information, including fraud detection, user habits, bottlenecks and security
breaches.

11.2.1 PROCESS OF DATA MINING — 4 STEPS

The Institute of Cost Accountants of India Page 2
CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

# Step Description
(i) Setting the The most difficult step. Data scientists and business stakeholders
Business Objective together identify the business challenge, which informs the data
queries and parameters for a specific project. Analysts may also
need to conduct further study to understand the company
environment.
(ii) Preparation of Once the scale of the problem is established, data scientists
Data determine which data collection will assist the company. The data
is then cleansed by eliminating any noise, such as repetitions,
missing numbers and outliers. Dimensionality reduction may also
be performed to maintain only the most essential predictors.
(iii) Model Building & Data scientists study intriguing relationships such as frequent
Pattern Mining patterns, clustering algorithms or correlations. Deep learning
algorithms may be used for classification (supervised learning
with labelled input data) or clustering (unsupervised learning with
unlabelled data).
(iv) Result Evaluation After aggregating data, findings must be analysed. Results must
& Implementation be valid, original, practical and comprehensible. When this
criterion is satisfied, companies can execute new strategies based
on this understanding to attain their intended goals.

11.2.2 TECHNIQUES OF DATA MINING

Technique Description
(i) Association Rule-based technique for discovering associations between variables
Rules inside a given dataset. Commonly employed for market basket analysis
— understanding linkages between various items. Helps create
effective cross-selling tactics and recommendation engines.
(ii) Neural Primarily used for deep learning algorithms. Replicates the
Networks interconnection of the human brain through layers of nodes to process
training data. Every node has inputs, weights, a bias (or threshold) and
an output. If output value exceeds a predetermined threshold, the node
'fires' and passes data to the subsequent layer. Learns using supervised
learning and gradient descent.
(iii) Decision Tree Uses classification or regression algorithms to classify or predict likely
outcomes based on a collection of decisions. Employs a tree-like
representation to depict the potential results of actions.
(iv) K-Nearest Classifies data points depending on their closeness and correlation with
Neighbour (KNN) other accessible data. Assumes similar data points exist in close
proximity to one another. Measures distance between data points (often
by Euclidean distance), then assigns the most common category or
average.

The Institute of Cost Accountants of India Page 3

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

11.2.3 IMPLEMENTATION OF DATA MINING IN FINANCE & MANAGEMENT

Application Description
(i) Detecting Money Money laundering is the illegal conversion of black money to
Laundering & Financial white money. Data mining techniques have advanced to detect
Crimes money laundering. The methodology provides a mechanism for
bank customers to detect or verify the detection of the anti-
money laundering impact.
(ii) Loan Repayment & Loan Distribution is the core business function of every bank.
Credit Policy Analysis The loan prediction system automatically computes the size of
the characteristics it employs and examines data pertaining to its
size. Data mining aids in management of all critical data and
massive databases by utilising its models.
(iii) Target Marketing Data mining and marketing work together to target a certain
market and assist in determining market decisions. With data
mining, it is possible to keep track of earnings, margins etc. and
determine which product is optimal for various types of
customers.
(iv) Design & The business is able to retrieve or move data into several huge
Construction of Data data warehouses, allowing a vast volume of data to be correctly
Warehouses and reliably evaluated with the aid of various data mining
methodologies and techniques.

11.3 Analytics and Model Building (Descriptive, Diagnostic, Predictive, Prescriptive)

Businesses utilise analytics to study and evaluate their data and translate their discoveries into insights
that aid executives, managers and operational personnel in making more educated and prudent business
choices.

Descriptive Diagnostic Predictive Analytics Prescriptive

Analytics Analytics WHAT could Analytics
WHAT happened? WHY did it happen? happen? WHAT should be
Historical data — Root cause analysis Future forecasting — done?
describes past — explores causes of predicts probable Action
events. outcomes. future outcomes. recommendations —
the most
sophisticated type.

DETAILED COMPARISON OF ALL FOUR ANALYTICS TYPES

The Institute of Cost Accountants of India Page 4

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

Type Key How it Works Examples Advantages /

Question Disadvantages
Descriptive What Uses two Company reports ADV: Easiest
Analytics happened? fundamental on inventory, type — uses
— What techniques: (1) Data workflow, sales simple
happened? aggregation — and revenue. maths/statistics
gathering and Social analytics (arithmetic,
organising data into (followers, likes, averages, %
digestible data sets. page visits). changes).
(2) Data Annual income Applicable to
mining/discovery — reports. Survey day-to-day
extracting patterns, data compilation. operations. Does
trends and not require in-
significance to depth analytics
present in an expertise. Quick
intelligible format. performance
5 Steps (Dan reporting.
Vesset): Step 1 — DIS: Does not
Decide business seek to go beyond
metrics; Step 2 — surface data.
Identify data Cannot draw
requirements; Step 3 inferences or
— Prepare and make forecasts.
collect data; Step 4
— Analyse data
(summary statistics,
clustering,
regression); Step 5
— Present data
using charts and
graphs.
Diagnostic Why did it Strategies include: HR department ADV: Derives
Analytics happen? data discovery, drill- examining value from data
— Why did it down, data mining employee by asking relevant
happen? and correlations. performance based questions and
Analysts identify on quarterly doing in-depth
data sources that performance levels, analyses. Turns
assist in interpreting absenteeism and data into visuals
outcomes. May need weekly overtime and insights.
to include external hours. Needs a BI and
data sources to find Cybersecurity team analytics platform
connections and determining the that is adaptable,
establish causality. relationship nimble and
Uses linear between security configurable.
regression to rating and number DIS: Requires
identify of incidents. both internal and
relationships. external data
sources. May
require multiple
datasets to
identify patterns
The Institute of Cost Accountants of India Page 5
CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

Type Key How it Works Examples Advantages /

Question Disadvantages
and establish
links.
Predictive What Foundation is E-commerce — ADV: Inventory
Analytics could probability. anticipating client forecasting;
— What happen? Techniques: data preferences based Customer service
could mining, statistical on previous personalisation;
happen? modelling purchases. Sales — Fraud detection
(mathematical estimating the and prevention;
relationships possibility that a Risk mitigation
between variables), buyer will (enhanced
machine learning purchase another applicant
algorithms item. Human screening in
(classification, resources — financial sector).
regression, identifying DIS: Based on
clustering). Machine employees probabilities —
learning algorithms contemplating can never be
utilise current data resignation. absolutely
and make best Healthcare — precise. Requires
feasible assumptions anticipating availability of
to fill in missing staffing historical data,
data. Deep learning requirements. typically in
is a recent subfield enormous
that imitates human quantities.
brain networks.
Prescriptive What Goes one step GPS technology — ADV: Gives
Analytics should be farther than gives important insights
— What done? descriptive and recommended for optimal data-
should be predictive analysis routes to driven decisions.
done? by advising the best destination based Significantly
potential business on travel time and minimises
actions. Uses rules, road closures. Oil likelihood of
statistics and and manufacturing human mistake.
machine learning — monitoring price Enables decision-
algorithms on both fluctuations. makers to assess
internal data and Healthcare — potential
external data (e.g., enhancing patient consequences
social media). care by analysing before making a
Predicts what, when, readmission rates. choice.
and why something Insurance — DIS: Requires
may occur and then evaluating enormous
recommends which customer risk. volumes of data.
options would best Machine learning
capitalise on future techniques cannot
opportunities or consistently
reduce future account for all
hazards. external variables.
Most
sophisticated —

The Institute of Cost Accountants of India Page 6

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

Type Key How it Works Examples Advantages /

Question Disadvantages
rarely used in
daily company
operations.

11.4 Standards for Data Tagging and Reporting (XML, XBRL)

Data tagging and reporting standards ensure that financial and business data is communicated in a
standardised, machine-readable format. The two most important standards are XML and XBRL.

XML — Extensible Markup Language XBRL — Extensible Business Reporting

Language

Definition Definition
A file format and markup language for A data description language that facilitates
storing, transferring and recreating arbitrary the interchange of standard, comprehensible
data. Specifies standards for encoding texts corporate data. It is based on XML and
in a format understandable by both humans enables automated interchange and
and machines. Defined by the 1998 XML trustworthy extraction of financial data
1.0 Specification of the World Wide Web across all software types and advanced
Consortium. technology, including the Internet.

Key Features Key Features

● Design objectives: Internet usability, ● Allows organisations to arrange data

universality and simplicity. using tags (e.g., 'revenue' tag means it

● Textual data format with significant pertains to revenue).
support for many human languages via ● Integrity of data is safeguarded by norms

Unicode. that have been already accepted.

● Primary function: Serialization — ● Offers expanded contextual information

storing, sending and rebuilding arbitrary on the precise data content of financial
data. documents.
● Labels, categorises and arranges ● XBRL tags may designate data as

information systematically using XML 'currency' or 'accounts' within a report.

tags. ● A business, a person or another software

● A 'well-formed' XML document programme may quickly produce a

complies to fundamental XML variety of output formats and reports
principles; a 'valid' document adheres to based on a financial statement.
its schema.
● XSD (XML Schema Definition) defines

required metadata for reading and

verifying XML.

Applications Benefits of XBRL

Exchange of data via the Internet. Document ● All reports are automatically created

formats: RSS, Atom, Office Open XML, from a single source of information —
OpenDocument, SVG, XHTML. reduces erroneous data entry and
Communication protocols: SOAP, XMPP. increases data reliability.

The Institute of Cost Accountants of India Page 7

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

XML — Extensible Markup Language XBRL — Extensible Business Reporting

Language

Message interchange format for AJAX. ● Reduces expenses by simplifying and

Industrial data standards: Health Level 7, automating the preparation and
OpenTravel Alliance, FpML, MISMO, production of reports for various clients.
National Information Exchange Model. ● Accelerates the decision-making of
financial entities such as banks and
rating services.
● Facilitates the publication of analyst and
investor reports.
● Access, comparison and analytic
capabilities for information are
unparalleled.

11.5 Cloud Computing, BI, AI, RPA and Machine Learning

11.5 Cloud Computing, Business Intelligence, AI, RPA and Machine Learning

11.5.1 CLOUD COMPUTING

WHAT IS CLOUD COMPUTING?

Cloud computing is the delivery of a variety of services through the Internet, or 'the cloud.' It
involves storing and accessing data via distant servers as opposed to local hard drives and
private datacenters. Before the advent of cloud computing, businesses had to acquire and
operate their own servers, necessitating the purchase of sufficient server capacity to
minimise the risk of downtime. Today's cloud service providers enable businesses to lessen
their reliance on costly onsite servers, maintenance staff and other IT resources.

THREE TYPES OF CLOUD COMPUTING

Type Description
(i) Private Cloud Offers a cloud environment exclusive to a single corporate
organisation. Physical components housed on-premises or in a vendor's
datacenter. High level of control since the private cloud is available to
just one enterprise.
Benefits: Customizable architecture, enhanced security procedures,
capacity to expand computer resources as needed.
Deployment: Business maintains private cloud infrastructure on-
premises (via intranet), or engages a third-party cloud service provider
to host and operate servers off-site.

The Institute of Cost Accountants of India Page 8

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

Type Description
(ii) Public Cloud Stores and manages access to data and applications through the
internet. Fully virtualized — shared resources may be utilised as
necessary.
Benefits: Enables enterprises to grow with more ease; option to pay for
cloud services on an as-needed basis (significant benefit over local
servers). Rigorous security measures prevent unauthorised access to
user data by other tenants.
(iii) Hybrid Cloud Blends private and public cloud models — enterprises exploit benefits
of shared resources while leveraging their existing IT infrastructure for
mission-critical security needs.
Benefits: Store sensitive data on-premises and access it through apps
hosted in the public cloud. Example: Keep sensitive user data in a
private cloud (for privacy compliance) and execute resource-intensive
computations in a public cloud.

Private Cloud Public Cloud Hybrid Cloud

Exclusive to one org Shared resources via internet Blends private + public
On-premises or vendor Pay-as-you-go model Sensitive data on-premises
datacenter Rigorous security measures Public cloud for heavy
Highest control & computation
customization

11.5.2 BUSINESS INTELLIGENCE (BI)

WHAT IS BUSINESS INTELLIGENCE?

Business intelligence includes business analytics, data mining, data visualisation, data tools
and infrastructure, and best practices to assist businesses in making choices that are more
data-driven. When you have a complete picture of your organisation's data and utilise it to
drive change, remove inefficiencies, and swiftly adjust to market or supply changes, you
have contemporary business intelligence. Modern BI systems promote adaptable self-
service analysis, controlled data on dependable platforms, empowered business users and
rapid insight delivery. History: Traditional BI originated in the 1960s as a method for
disseminating information across enterprises. The phrase 'Business Intelligence' was coined
in 1989.

9 BI METHODS

BI Method Description
(i) Data Mining Large datasets may be mined for patterns using databases,
analytics and machine learning (ML).
(ii) Reporting The dissemination of data analysis to stakeholders in order for
them to form conclusions and make decisions.

The Institute of Cost Accountants of India Page 9

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

BI Method Description
(iii) Performance Comparing current performance data to previous performance
Metrics & data in order to measure performance versus objectives,
Benchmarking generally utilising customised dashboards.
(iv) Descriptive Utilizing basic data analysis to determine what transpired.
Analytics
(v) Querying BI extracts responses from data sets in response to data-specific
queries.
(vi) Statistical Analysis Taking the results of descriptive analytics and using statistics to
further explore the data, such as how and why this pattern
occurred.
(vii) Data Visualisation Data consumption is facilitated by transforming data analysis
into visual representations such as charts, graphs and histograms.
(viii) Visual Analysis Exploring data using visual storytelling to share findings in real-
time and maintain the flow of analysis.
(ix) Data Preparation Multiple data source compilation, dimension and measurement
identification, and data analysis preparation.

11.5.3 ARTIFICIAL INTELLIGENCE (AI)

DEFINITION OF AI
John McCarthy of Stanford University defined Artificial Intelligence as: 'It is the science and
engineering of making intelligent machines, especially intelligent computer programs. It is
related to the similar task of using computers to understand human intelligence, but AI does
not have to confine itself to methods that are biologically observable.' Alan Turing's
landmark paper 'Computing Machinery and Intelligence' marked the genesis of AI discourse.
Turing, the 'father of computer science,' posed the question 'Can machines think?' and
proposed the famous Turing Test — where a human interrogator attempts to differentiate
between a machine and a human written answer. In its simplest form, AI combines
computer science and substantial datasets to allow problem-solving. It includes the subfields
of machine learning and deep learning.

HUMAN APPROACH (RUSSELL & IDEAL APPROACH (RUSSELL &

NORVIG) NORVIG)
Systems that think like humans Systems that think rationally
Systems that act like humans Systems that act rationally

AI APPLICATIONS IN FINANCE (FIG 11.2)

INVESTMENT SERVICES LENDING

✓ (i) Algorithmic Trading

The Institute of Cost Accountants of India Page 10

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

✓ (ii) Robo Advisory ✓ (i) Retail and Commercial Lending

✓ (iii) Insurance Claim Processing Operations
✓ (iv) Pricing of Insurance Products ✓ (ii) Retail and Commercial Lending
Scores
✓ (iii) Detecting Possibility of Default

ARTIFICIAL INTELLIGENCE IN
FINANCE

AUDIT AND COMPLIANCE CUSTOMER SERVICE

✓ (i) Fraud Detection ✓ (i) Detecting New Sell Opportunities
✓ (ii) Regulatory Compliance ✓ (ii) Know Your Customer (KYC)
✓ (iii) Travel and Expense Management ✓ (iii) Prediction of Customer Churning

TYPES OF AI — WEAK AI VS STRONG AI

WEAK AI (Narrow AI / ANI) STRONG AI (AGI + ASI)

Also known as: Artificial Narrow Comprises: AGI (Artificial General

Intelligence (ANI) Intelligence) + ASI (Artificial Super
AI that has been trained and honed to do Intelligence)
particular tasks. Most AI that surrounds us AGI: A hypothetical kind of AI in which a
today is powered by weak AI. machine possesses human-level intellect,
Examples: self-aware consciousness, and the ability to
solve problems, learn and plan for the
Apple's Siri | Amazon's Alexa | IBM
future.
Watson | Driverless cars
ASI (Superintelligence): Would transcend
the intelligence and capabilities of the
human brain.
Status: Still totally theoretical — no
practical applications yet. Sci-fi example:
HAL from 2001: A Space Odyssey.

DEEP LEARNING VS MACHINE LEARNING — KEY DIFFERENCES

Both deep learning and machine learning are subfields of AI. However, deep learning is a subfield of
machine learning. The hierarchy is: AI → Machine Learning → Deep Learning (innermost).

Machine Learning (ML) Deep Learning (DL)

A subfield of Artificial Intelligence. A subfield of Machine Learning (which is

Classical or 'non-deep' ML requires more itself a subfield of AI).
human interaction to learn. 'Deep' = neural network with more than
Human specialists develop the hierarchy of three layers (inputs + multiple hidden layers
characteristics to understand distinctions + outputs).
between data inputs.

The Institute of Cost Accountants of India Page 11

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

Machine Learning (ML) Deep Learning (DL)

Often requires more structured data to learn. Automates feature extraction — reduces the
need for manual human involvement.
Can ingest unstructured data in raw form
(text, images) without human interaction.
Automatically establishes the hierarchy of
characteristics. Also known as 'scalable
machine learning' (Lex Fridman).

Can use supervised (labelled) or Can use labelled datasets (supervised

unsupervised learning. learning) or unlabelled data — not
required.

11.5.4 ROBOTIC PROCESS AUTOMATION (RPA)

WHAT IS RPA?
With RPA, software users develop software robots or 'bots' that are capable of learning,
simulating and executing rules-based business processes. By studying human digital
behaviours, RPA automation enables users to construct bots. Robotic Process Automation
software bots can communicate with any application or system in the same manner that
humans can — but operate around the clock without fatigue, errors or breaks.

Key Points about RPA

RPA bots simulate and execute rules-based business processes by studying human digital
behaviours.
Bots can be given instructions and then left to complete the task autonomously.
RPA bots can communicate with any application or system in the same manner that humans
can.
Unlike human workers, RPA bots operate 24/7 without fatigue, errors or breaks.
RPA is used to automate highly repetitive, routine tasks normally performed by knowledge
workers.
Applications in finance: Invoice processing, data entry, reconciliation, compliance reporting,
payroll processing.
RPA differs from AI: RPA follows pre-defined rules, while AI can learn and adapt
dynamically.

RPA Bots — Digital Workforce

RPA bots possess a digital skill set that exceeds that of humans. Consider RPA bots to be a
Digital Workforce capable of interacting with any system or application.
Bots can: copy-paste, scrape site data, do computations, access and transfer files, analyse
emails, log into programmes, connect to APIs, and extract unstructured data.
Due to the adaptability of bots to any interface or workflow, there is no need to modify
existing corporate systems, apps or processes in order to automate.

The Institute of Cost Accountants of India Page 12

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

RPA bots are simple to configure, utilise and distribute — as easy as pressing record, play
and stop buttons and using drag-and-drop.
RPA bots may be scheduled, copied, altered and shared to conduct enterprise-wide business
operations.
RPA bots function continuously, around-the-clock, with 100 percent accuracy and
dependability.

7 BENEFITS OF RPA

Benefit Description
(i) Higher RPA bots operate continuously around-the-clock, 24/7, without
Productivity breaks or fatigue — dramatically increasing overall output.
(ii) Higher Accuracy Bots function with 100% accuracy and dependability, eliminating
human errors in repetitive tasks.
(iii) Saving of Cost Automating repetitive, labour-intensive tasks reduces operational
costs and frees up human resources for higher-value work.
(iv) Integration Bots can communicate with any application or system — no need to
Across Platforms modify existing corporate systems, apps or processes in order to
automate.
(v) Better Customer Faster processing and fewer errors lead to improved service delivery
Experience and a better experience for customers.
(vi) Harnessing AI RPA can be combined with AI capabilities (such as NLP and ML) to
handle more complex, cognitive tasks beyond simple rule-based
automation.
(vii) Scalability RPA bots can be scheduled, copied, altered and shared to conduct
enterprise-wide business operations — easily scaled up or down as
business needs change.

11.5.5 Machine Learning (ML)

WHAT IS MACHINE LEARNING?

Machine learning (ML) is a branch of study devoted to understanding and developing
systems that 'learn,' or ways that use data to improve performance on a set of tasks. It is
considered a component of Artificial Intelligence. Machine learning algorithms construct a
model based on training data and sample data in order to generate predictions or conclusions
without being explicitly taught to do so. Applications: Medicine, email filtering, speech
recognition, computer vision — areas where it is difficult or impractical to create traditional
algorithms. Key Premise: Tactics, algorithms and conclusions that performed well in the
past are likely to continue to perform well in the future. Programs capable of ML can
complete tasks without being expressly designed to do so — they learn from available data.
For complex jobs, it may be more efficient to assist the computer in developing its own
algorithm rather than having human programmers describe each step.

The Institute of Cost Accountants of India Page 13

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

5 APPROACHES TO MACHINE LEARNING (BASED ON TYPE OF SIGNAL /

FEEDBACK)

(i) Supervised (ii) (iii) Semi- (iv) (v) Self-

Uses labelled Unsupervised Supervised Reinforcement Supervised
training data Uses unlabelled Combines small Agent learns Data itself
with inputs + data to discover labelled data from rewards provides
expected structure via with large and penalties by supervision —
outputs. grouping & unlabelled data. interacting with no external
clustering. environment. annotation
needed.

DETAILED DESCRIPTION OF ALL 5 MACHINE LEARNING APPROACHES

Approach Description
(i) Supervised Algorithms construct a mathematical model of a dataset that includes
Learning both inputs and expected outcomes (training data). Each training
example has one or more inputs and the expected output (supervisory
signal). By optimising an objective function iteratively, the algorithm
discovers a function to predict outputs for new inputs.
Key concepts: Feature vector, training data matrix, objective function
optimisation.
Sub-types: Classification (outputs limited to a certain set of values,
e.g., email spam filter), Regression (outputs may take any value
within a range), Similarity learning (quantifies how similar/related
two items are — used in ranking, recommendation systems, face
verification).
(ii) Unsupervised Uses a dataset comprising just inputs (unlabeled, unclassified and
Learning uncategorised test data) to identify data structure such as grouping
and clustering. Algorithms identify similarities in the data and
respond based on the presence or absence of such similarities in each
new dataset.
Applications: Density estimation (calculating probability density
function), data feature summary and explanation.
Cluster analysis: Assigning a set of data to subsets (clusters) so that
observations within the same cluster are similar, while observations
from other clusters are different.
(iii) Semi- Intermediate between unsupervised learning (without labelled
Supervised training data) and supervised learning (with completely labelled
Learning training data). When unlabeled data is combined with a small
quantity of labelled data, there is a significant gain in learning
accuracy.
In poorly supervised learning, training labels are noisy, restricted or
inaccurate; yet, these labels are frequently less expensive to acquire,
resulting in larger effective training sets.

The Institute of Cost Accountants of India Page 14

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

Approach Description
(iv) Reinforcement The learning system (agent) interacts with an environment, observing
Learning the results of its actions and learning from rewards or penalties. The
goal is to maximise cumulative reward over time.
Key feature: The agent learns from trial and error — it is not told
which actions to take but must discover which actions yield the most
reward.
Applications: Game playing (AlphaGo), robotics, algorithmic
trading, autonomous vehicles.
(v) Dimensionality Dimensionality reduction is the process of acquiring a set of major
reduction variables in order to reduce the number of random variables under
consideration. In other words, it is the process of lowering the size of
the feature set, which is also referred to as the “number of features.”

SUPERVISED LEARNING — SUB-TYPES

Classification Regression Similarity Learning

Outputs are limited to a Outputs may take on any Quantifies how similar or
certain set of values. value within a given range. related two items are.
Example: Email spam filter Example: Predicting house Applications: Ranking,
— input is an email; output prices, stock prices, sales recommendation systems,
is the folder name. forecasting. face verification, speaker
verification.

NOTE
This section covers: Cloud Computing, Business Intelligence (BI), Artificial Intelligence
(AI), Robotic Process Automation (RPA) and Machine Learning — and their applications in
business and finance. These topics are covered in the complete ICAI textbook (pages not
included in the current upload). Please refer to your ICAI study material for the full content
of Section 11.5.
Key Terms to Know:
◆ Cloud Computing: Delivery of computing services (servers, storage, databases,
networking, software) over the internet ('the cloud') to offer faster innovation, flexible
resources and economies of scale.
◆ Business Intelligence (BI): Technologies, applications and practices for the collection,
integration, analysis and presentation of business information to support better business
decision making.
◆ Artificial Intelligence (AI): Simulation of human intelligence processes by computer
systems, including learning, reasoning and self-correction.
◆ Robotic Process Automation (RPA): Use of software robots or 'bots' to automate highly
repetitive, routine tasks normally performed by knowledge workers.
◆ Machine Learning (ML): A subset of AI — a computer system's ability to learn from
and adapt to data without being explicitly programmed.

The Institute of Cost Accountants of India Page 15

CMA Inter | Module 11 | Data Analysis and Modelling Paper 11 | Section B

11.6 Model vs. Data-driven Decision-making

Model-driven vs. Data-driven Decision Making

MODEL-DRIVEN: Decision making based on pre-built mathematical or statistical models.
The model represents a simplified abstraction of reality. Managers use the model's output as
the primary input for decisions. Examples: NPV models, DCF models, linear programming.
DATA-DRIVEN: Decision making based on data analysis and interpretation. Decisions are
made based on actual data rather than assumptions or intuition. Requires robust data
collection, cleaning, analysis and visualisation infrastructure. Examples: Real-time analytics
dashboards, machine learning-based recommendations.
KEY DIFFERENCE: Model-driven approaches start with a model and fit data to it; data-
driven approaches start with data and derive patterns/models from it.
NOTE: The complete content for this section is available in the ICAI textbook. Please refer
to Section 11.6 in your study material for the full discussion.

The Institute of Cost Accountants of India Page 16