Supervised

vs Unsupervised Learning

 [Link]
 Content
• Machine Learning
• Supervised Learning
• Unsupervised Learning
• ML Applications

 [Link]
 Learning Objectives
• Why Machine Learning?
• What is Machine Learning?
• What is Supervised Learning?
• Applications of Supervised Learning?
• What is Unsupervised Learning?
• Applications of Unsupervised Learning?

 [Link]
 Why Machine Learning?
• Why Machine Learning?
• Everyone like to know the Future
• Adapt and learn fast with changing scenario
• Act fast with changing data

• What is Machine Learning?

• An algorithm that learns from data, identifies
patterns in data and store the learnings in
form of a Model
• Apply the Model to predict on new data
• It has the ability to quickly change, refresh,
and enhance the Model with changing data
and newer datasets

 [Link]
 Simple Business Scenario
Scenario
Let us assume you are working in a Bank and the Chief Marketing
Officer suggests that he wish to run a campaign to promote a financial
product, say, some Investment Product

Based on business filters, you have an eligible contactable base of

1,000,000 customers.
Cost of Targeting each customer being Rs. 10/-
It is expected that 0.5% incremental customers will purchase the
Investment Product because of the campaign
Expected Revenue per customer who purchases the product is Rs.
2500/-

 [Link]
 Campaign Return on Marketing Investment
without Analytical Approach
• Target Customer Base : 1,000,000
• Cost of Targeting per customer : INR 10/-
• Cost of Campaign = 1,000,000 * 10 = INR 10,000,000 = 10 Mn
• Expected Incremental Conversion Rate : 0.5%
• Expected Incremental Conversions = 1,000,000 * 0.5% = 5,000
• Expected Revenue per Convert : INR 2500/-
• Expected Incremental Revenue = 5,000 * 2500 = 12,500,000 = 12.5
Mn
• Expected Profit = 12.5 Mn – 10 Mn = 2.5 Mn

 [Link]
 Campaign ROMI

Return on Revenue – Cost 12.5 - 10

Marketing = ----------------------- = -------------------- = 25%
Investment (ROMI) Cost 10

 [Link]
 Analytics Based Approach

High Response Segment

25% of Base
With expected conversion rate of 1.3%

Medium Response Segment

25% of Base
With expected conversion rate of 0.4%

Low Response Segment

50% of Base
With expected conversion rate of 0.15%

 [Link]
 Analytics Based ROMI
Note: Cost of Targeting per customer : INR 10/- ; Expected Revenue per Convert : INR 2500/-

Exp. Conv. Cost of Exp.

# Customer # Conv’s Profit ROMI
Rate Targeting Revenue
Segment
(E = C *
(A) (C = A * B) (F = E – D) G=F/D
(B) (D = A * 10) 2500)
High
Response 250,000 1.3% 3250 2,500,000 8,125,000 5,625,000 225%
Segment
Medium
Response 250,000 0.4% 1000 2,500,000 2,500,000 0 0%
Segment
Low
Response 500,000 0.15% 750 5,000,000 1,875,000 -3,125,000 -ve
Segment

Total 1,000,000 0.5% 5000 10,000,000 12,500,000 2,500,000 25%

 [Link]
 Recommendation to CMO

Your recommendation to the CMO:

• Target only the High Response Segment

Benefits of your strategy

A) It will reduce Marketing Cost by 75%

B) It will increase Profits by 125%

C) 9X increase in ROMI

 [Link]
 Supervised
vs Unsupervised Learning

 [Link]
 Machine Learning Techniques Categories
 Supervised learning is the Machine Learning task of
finding a function from a Labeled Data
 Labeled Data is a dataset which has Independent
Variable/s and a Dependent Variable

 Unsupervised learning is the Machine Learning task

of exploring the data to derive some inferences /
insights from the dataset
 The “Target Variable” or the “Labeled Class” is not
present in the Unsupervised Learning dataset

 [Link]
 Supervised Learning

Input
D Attributes Supervised
A
Learning
Predictive Predicted
T
A Desired Technique Model Output
Output

• Supervised Learning Techniques

• Classification
• Regression

 [Link]
 UnSupervised Learning

Input Unsupervised

Data
Learning Output
Technique

• UnSupervised Learning Techniques

• Dimension Reduction Techniques like PCA, Factor Analysis
• Clustering
• Association Analysis

 [Link]
 Application of Supervised Learning
• Assume you are working in a bank (say MyBank)
• The Chief Marketing Officer has assigned you the task of
growing the Personal Loans Portfolio by cross-selling the
loans to existing Customers
• Data of past promotional campaigns and offers sent to the
Customers, their behavioural data and those who took the
loan is all available with you

• This is an example where Supervised Learning can be

applied

 [Link]
 Marketing Modeling Dataset
• Sample Predictive Modeling Dataset

 [Link]
 Some e.g. of Supervised Learning Applications
Industry /
Supervised Learning Technique Applications Labeled Class
Vertical

HR To predict whether a good employee is likely to resign or not Resign / Not-Resign

Telecom To classify customers who are likely to be Churners Churn / Not-Churn

Retail / To find potential customers from churned base who can be won back Win-back Yes / No
Ecommerce again

Banking To build a model that will help assign the probability to a customer to Respond / Not-
take a product / service Respond

Insurance To build a model to assess the likelihood of customer not renewing his / Lapse / Not-Lapse
her policy

 [Link]
 Application of UnSupervised Learning
• Assume you are working in a Retail Company
• You have 1 Mn Loyalty Members
• You have been asked to segment them based on their Buying
Behaviour Pattern

• This is an example of UnSupervised Learning Application

 [Link]
 Clustering Modeling Dataset
• Sample Clustering Modeling Dataset

 [Link]
 Thank you

Contact us:
[Link]@[Link]

 [Link]