Techconnex Think beyond BI: Machine Learning

enabling modern enterprise
1
Think Beyond
Business Intelligence
for your organization –
Machine Learning
Albert Hui, MBA, M.A.Sc., CSM
Data Economist

2/4/2016 2enabling modern enterprise
About Me
2
• Co Founder of Data Economist, a data consulting firm based in Toronto.
• 18 years in data management consulting, business Intelligence, data
warehousing and big data.
• Master in Engineering in the area of Artificial Intelligence
• Big Data Architect, Data Scientist
• Conference Speaker at IOUG, TOUG Collaborate since 2011
• Technical editor on Oracle 12c Book.
• M.A.Sc., MBA, University of Toronto
• Toronto based
• Twitter: @dataeconomist
• Father of two twin boys

Agenda
3
Objective of this Session
What is machine learning?
BI vs ML
Models and Tools
Use Cases
What’s for the Organization?
Concluding Thoughts

Objective of this Talk
4
Introduction of Machine Learning

5
History of Data &
Machine Learning

The Meaning of.....
6
 Data:
 Granular and raw and viewed as the
lowest level of abstraction from which
information and knowledge are derived.
 Information:
 Extracting data in order to effectively
derive value and meaning and
establishing a relevant context, often
selecting from many possible contexts.

The Meaning of.....
7
 Intelligence:
Individuals differ from one another in their
ability to understand complex ideas, to
adapt effectively to the environment, to
learn from experience.
 Wisdom:
A deep understanding of people, things,
events or situations, resulting in the ability
to choose or act to consistently produce the
optimum results with a minimum of time
and energy.

8
Gandalf
Legolas

Build
Knowledge
thru
Experience
9

Definition
11
Computer systems that
automatically learn or
improve with experience
(data).

What is machine learning?
12
Machine Learning is not new

Common Machine Learning Applications
13

Machine learning: two major types
14
 Supervised
 Supervised learning is tasked with learning a function from labelled
training data in order to predict the value of any valid input. Common
examples of supervised learning include classifying e-mail messages
as spam, labelling Web pages according to their genre, and
recognizing handwriting. Many algorithms are used to create
supervised learners, the most common being neural networks,
Support Vector Machines (SVMs), and Naive Bayes classifiers.
 Unsupervised
 Unsupervised learning, is tasked with making sense of data without
any examples of what is correct or incorrect. It is most commonly
used for clustering similar input into logical groups. It also can be
used to reduce the number of dimensions in a data set in order to
focus on only the most useful attributes, or to detect trends. Common
approaches to unsupervised learning include k-Means, hierarchical
clustering, and self-organizing maps

Data mining vs machine learning
15
 Data Mining
 Focus on extracting patterns, unknown properties on the data.
 Marketing
 Surveillance
 Fraud Detection
 science discovery
 Discover items usually purchased together
 Machine learning
 Focus on extracting prediction models, based on known
properties learned from the training data
 E-Mail spam classification
 News-topic discovery
 Building recommendations

16
“I skate to
where the puck
Is going to be,
not where it
has been”
-- Wayne Gretzky

Quick Quiz
18
 In US, a 45year male, Around 150-180K
income, Post Graduate Education, if he wants
to buy a car. Which brand?

Quick Quiz
19
 In US, a 45year male, 3 children, 180K
income, Graduate School Education, if he
wants to buy a car. And he lives in Texas,
then which brand?

Quick Quiz
20
 In US, a 45year male, 3 children, Graduate
School Education, making $60K/year. If he
wants to buy a car. Then which brand?

Remember this?
22

23
Make a decision

BI Reports ….
24

Key factors of Successful
Enterprises25

26

Learning Enterprise
27

28

Decide – Learn -
Experience29
Make better Builds

How do we keep up
changes?30
When the winds of
change blow,
some people build
walls and others
build windmills
-- Chinese proverb

31

Main Models
33
Classification Regression
Clustering

Machine Learning Components
34

Machine Learning : Data
35

Modelling Concept (Dimensionality Reduction)
Sales
Project three dimensional space
into two dimensional space
Principal Component Analysis

Modelling Concept (generalization)
37
x
y
Is it a better model?

3
1
5
2
5
4
4
3
32
210 xxbxb
x
xb
xbxbby
)min(
0
 

ii
n
i
b
Objective Function
Regularization

Build the model
38
 Learn a model from a manually trained dataset
 Predict the class of an unseen object based on
features

Build the model: iterative process
39
 There is no single answer/model to your
questions.
 It is often based on trial and error.

Some Models Available
40
 Multiple Variable Regression
 Decision Tree/Random Forrest
 Logistic Regression
 Neural Network
 Fuzzy Logic
 Support Vector Machine
 Bayesian Network
 K-means
 KNN – Knearest Neigbors

Tools
41

42
Use Case #1
Ad Click-thru rate
Prediction using
Machine Learning

43

Clickstream – Case Demo
44
 An Asia based Hotspot Wi-Fi provider
 Revenue Model: Advertising
 Advertisers place ads before users can connect to Wifi
 Data
 Survey data: Users are required to fill a survey before
logging in.
 Click logs including Ad click-through
 Data Size:
 12GB+ compressed a day.
 15M signed on and 6% click-thru a day.
 Problem definition: click-through rate is too low
Demo

Preference Matching :
Clustering45
Matching
Millions of People Thousands of Ads

Logistic Regression
46

Ad: n-dimensional vectors
 Each Ad is represented as a vector of
customers who have clicked the ads.
 Probability of Ad clicked based on how
close with individual Ad vectors.
Vector for customers who clicked and viewed burger king
Vector for customers who clicked and viewed wholefood
Vector for an signed on customer

Improvement
48
Click-thru rate increased from 6% to
13% first year and 19% next year.

49
Use Case #2
A Major Canadian
retailer

Data Science – First Question
50
What are the customers saying?

Listen to Customers
51
install.packages("twitteR")
rdmTweets <- userTimeline("rdatamining", n=100)
install.packages("ROAuth")
install.packages("RCurl")
install.packages("tm")
install.packages("SnowballC")
install.packages("wordcloud")
twitCred <- OAuthFactory$new(consumerKey="rp4fHagxYdsdfoSMR3FiOg",
consumerSecret="webKiy93XZZDLQcBJzxHw64regfgZvbivJIpLctYcPKY",
requestURL="https://api.twitter.com/oauth/request_token",
authURL="https://api.twitter.com/oauth/authorize",
accessURL="https://api.twitter.com/oauth/access_token")
wordcloud(d$word, d$freq, min.freq=3)

NPL : Sentiment Analysis
52
weather
waiting
Olympics

2/4/2016 53enabling modern enterprise53
Words Popularity
Olympics 0.9
Winter 0.9
Customer Service 0.85
Computer 0.85
Annoy 0.85
Auto Oil 0.7
horrible 0.7
line 0.7
wait 0.7
Found 0.65
traffic 0.6
walmart 0.6
good 0.6
back 0.57
Target 0.56
Parking 0.56
sick 0.55
Association Association
Game thanks
Olympics weather
wait product
wait slow
Asile wait
Engine price
wait stand
stand customer Service
time find
line stand
target weather
phone find
Olympics family
worst good
find parking
target find
service slow
dictCorpus <- myCorpus
# stem words in a text document with the snowball stemmers,
# which requires packages Snowball, RWeka, rJava, RWekajars
myCorpus <- tm_map(myCorpus, stemDocument)
# inspect the first three ``documents"
inspect(myCorpus[1:3])
myCorpus <- tm_map(myCorpus, stemCompletion, dictionary=dictCorpus)
inspect(myCorpus[1:3])
myDtm <- TermDocumentMatrix(myCorpus, control = list(minWordLength = 1))
#inspect(myDtm[266:270,31:40])
print(myDtm)
findFreqTerms(myDtm, lowfreq=2)
findAssocs(myDtm, 'Olympics', 0.10)
findAssocs(myDtm, 'service', 0.10)
NPL : Sentiment Analysis

Improvements
54
 Improve the
customer service line
 Increase Customer
Service Staff on Sat.
 Reduce Wait Time

Data Science – What Impact Sales
55
Product
Sales
Weather
Store
Demographic
Product
Review
Inventory
On-line
Clickstream
Competitors
Store Size &
Attributes
Product Price
Environmental Data
Environic Data
Site, Forum
Review, twitter
Store Inventory
Site Clickstream
Marketing
Analysis
Dealer
Management
Promotion

Data Science – What Impact Sales
56
Multivariate Linear Regression

57
Use Case # 3
Cash flow projection
for a major bank

Deposit Saving : Asset Liability Modeling
58
Asset Allocation Optimization
1. Predicting cash balance for
customer segments.
2. Optimizing lending and asset
allocation.
3. Minimize liquidity risk.
4. Enhance pricing strategy.
5. Manage better customer
relationship.
Every 10K deposit
10d 20d 1m …... 6m 1y 2 y
Survival model
for Cash flow projection
based on customer
profile
Retirees 65+
Single Saver
40 with kids
Single spender

Use Case #3: Benefits
59
Estimated 100+ Millions
per year revenue
opportunity

60
What’s for your
organization?

Enhance your BI Reports with prediction
capability61
Provide
predictive
capability in your
BI Reports.

Decision Architecture
62
models
business
applications
BI Reports
Feedback

Decision as a Service
63
Make better
Builds

Concluding Thoughts:
64
In this connected age,
what is the most
disruptive?

Customers are actually more disruptive than
technology

Decisions Made by IoT
To create the experience,
decisions need to
be made where the events
happen.

Thank you!
67
Albert Hui, MBA, MASc., P.Eng, CSM
Data Economist Inc.,
Email: albert_hui@dataeconomist.com
Follow me at Twitter: @dataeconomist

Techconnex Think beyond BI: Machine Learning

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to Techconnex Think beyond BI: Machine Learning

Similar to Techconnex Think beyond BI: Machine Learning (20)

Recently uploaded

Recently uploaded (20)

Techconnex Think beyond BI: Machine Learning