Predictive Analytics-I
Vishnuprasad Nagadevara
Indian Institute of Management Bangalore
nagadev@[Link]
Schedule
Date Topics to be covered
12-Jun Introduction to data mining
12-Jun Introduction to text mining;
13-Jun Nave Bayes method for classification
13-Jun k-nearest neighbor algorithm
19-Jun Support vector machines
19-Jun Ensemble of classifiers: Bagging and Boosting
20-Jun Cluster analysis Data standardization, Hierarchical clustering
20-Jun K-means clustering; Cluster evaluation
26-Jun Text mining: text and web-page preprocessing
26-Jun Latent semantic indexing
Opinion mining and sentiment analysis: Document sentiment
27-Jun
classification, Aspect-based opinion mining
Schedule
Date Topics to be covered
Opinion mining and sentiment analysis: Mining comparative opinions;
27-Jun
Further topics
31-Jul Introduction to software to be used in the subsequent sessions:
Social network analysis: Co-citation and bibliographic coupling,
31-Jul
PageRank
Social media and network analysis: HITS algorithm, Community
01-Aug
discovery
01-Aug Introduction to web-mining, Web structure mining
07-Aug Web content mining
07-Aug Web usage mining
08-Aug Project presentation (Two sessions)
Project Proposals: Stage 1 due on 27 Jun
Term Paper Due on 1 Aug
Evaluation
Term Paper (40%)
Project (40%)
Project Report (30%)
Presentation (10%)
Final Exam (Take Home) (20%)
Term Paper
Select one of the following Three
A Specific Data Mining Technique (Eg. CART)
A Specific Industry (Eg. Banking or Insurance)
A Specific Application (Eg. Customer Churn)
Carry out Desk Research on various
applications with respect to the other two
dimensions
Credit for original critique and not for Cut
& Paste ability
Data Mining Models - Applications
CRM Customer acquisition, retention and upselling
Health Predict Continuation of drug usage
Quality Control Silk, Railways, Printing
Predicting Election Results
Fraud Detection Insurance, Income Tax, Customs
Predicting Stock Price Movements
Software Improving software quality
Security Protecting against Intrusions
Analytical CRM
Analytical CRM
Statistical methods for identifying the most (potentially)
profitable customers and prospects;
Strategic CRM
A strategic approach to acquiring, retaining and
developing customers to improve profitability in
exchange for superior value to the customer;
Operational CRM
Day-to-day processes to acquire profitable customers and
gain the highest degree of customer share while making
marketing, sales and service effective and productive
Customer Lifecycle
Management
Purpose is to Envelop the Customer so
Deeply in the Corporate Mesh that they are
Captured and Retained for Eternity
Customer Lifetime Value
CLV is the Measurement of the Expected
Cash flow, Gross Revenue and Margin
Contribution to Revenue over the Life-Time
of an Individual Customer
CLV Example
Expected Revenue from a New Retention
Customer Rate 60.00%
Year 1 Year 2 Year 3 Year 4
New Customers 1,000
Expected No. of Customers 600 360 216
Average Sale for Each Customer 10,000 10,000 10,000 10,000
Revenue 10,000,000 6,000,000 3,600,000 2,160,000
Expected Revenue from New
Customers 10,000,000 16,000,000 19,600,000 21,760,000
Average Revenue from Each of the
New Customers 10,000 16,000 19,600 21,760
CLV Example
Expected Profit from a New Retention
Customer Rate 60.00%
Year 1 Year 2 Year 3 Year 4
New Customers 1,000
Expected No. of Customers 600 360 216
Average Profit for Each Customer 1,000 1,000 1,000 1,000
Profit 1,000,000 600,000 360,000 216,000
Expected Profit from New
Customers 1,000,000 1,600,000 1,960,000 2,176,000
Average Profit from Each of the
New Customers 1,600 1,960 2,176
Use Discount rate 1,000,000 480,000 230,400 110,592
Cumulative 1,000,000 1,480,000 1,710,400 1,820,992
Per Customer 1,000 1,480 1,710 1,821
Discount Rate 25.00%
Customer Life-time Value
7000
6000
5000
4000
NPV
3000
2000
1000
0
0
5
0
5
0
5
0
5
0
5
5
5
6
6
7
7
8
8
9
9
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
Retention Rate
The Lift Chart
The Lift Chart
250
200
150
Hits
100
50
0
0 2000 4000 6000 8000 10000
Records Selected
The Lift Chart
The Lift Chart
250
200
150
Hits
100
50
0
0 2000 4000 6000 8000 10000
Records Selected
Customer Value Segments
25-Jun-17
Case of Moving Up
Total Per Cust
No. of Customers 2,130
Revenue 9,054,000 4,251
Profit 858,000 403
Margin 9.48% 9.48%
Case of Moving Up
Range Number Percentage Revenue %Revenue $ Rev $/Cust
From To
Top more than 58,900 21 0.99% 24.00% 2,172,960 103,474
Big 19,000 58,900 85 3.99% 27.00% 2,444,580 28,760
Medium 4,820 19,000 319 14.98% 29.00% 2,625,660 8,231
Small 1 4,820 1,705 80.05% 20.00% 1,810,800 1,062
All 2130 9054000
Case of Moving Up
Proft/ Move up
Number Cust % Profit $ 2% New New Rev New Profit
Top 21 33,949 712,929 23 2,379,909 780,827
Big 85 7,598 645,830 2 89 2,559,619 676,222
Medium 319 1,220 389,180 6 347 2,856,125 423,340
Small 1,705 -522 -890,010 34 1,671 1,774,690 -872,262
All 2130 857929 2130 9,570,343 1,008,127
Change 0% 516,343 150,198
Percent 0% 5.70% 17.51%
What is Data Mining
Process of Discovering Meaningful New
Correlations, Patterns and Trends through
Large Amounts of Data Using Pattern
Recognition Technologies as well as
Statistical and Mathematical Techniques
What is Data Mining
Attempts to Discover Knowledge Buried
in Databases
Describes a variety of Data Processing
Tools
Automated Discovery Process at an
advanced level
Different Nomenclature
Knowledge Discovery
Business Intelligence
Analytical Services
CRM
Examples of Data mining?
Direct Marketing
Customer Acquisition
Customer Retention
Cross Selling
Fraud Detection
Trend Analysis
Why Data Mining?
Who is my most profitable customer?
Why will my customer jump to my
competitor?
Which product will my customer buy, in
addition to my existing products?
History of DM
Drivers of DM
Statistics and DM
DM Uses Powerful Statistical Techniques
Statistics is for Confirming Theories
DM Does not make any Pre-judgment
Combination of Statistical Methods and
Validation Techniques Using Data
Warehouse
Query and Data Mining
Query:
Is there a battleship in Row B, Column 3?
Data Mining:
What are the most likely positions where the
opponent would place his Battleship?
DM for Competitive
Advantage
Problem Definition
Identify profitable groups to predict how to
retain them
Identify profitable group to predict how to cross
sell to them
Discovery
Discover more about your customers
Who are most important, most profitable
(Power Users)?
DM for Competitive
Advantage
Implementation
Create campaigns for the power users
DM Determines different offers to different
segments
Taking Action
Identify customer touch points
Campaign targeting using DM models
DM for Competitive
Advantage
Monitor Results
Generate metrics
Evaluate effectiveness of DM models
Incorporate Dynamic Learning
Algorithms
Cluster Analysis
Classification Trees
Neural Networks
SVM
Associations
Discriminant Analysis
Genetic Algorithms
Logistic Regression
Questions
Clarifications?