This presentation was done by Dinesh Priyankara (Senior Architect - Specialist at Virtusa) at the SLASSCOM Tech Talk - 'Smart Data Engineering' on 26th November 2014.

1. www.cammanagementsolutions.com
MACHINE LEARNING
Dinesh Priyankara
Senior Architect - Virtusa
MVP – SQL Server
http://dinesql.blogspot.com/

2. WE ARE ALREADY EXPERIENCING……
• Your bank calls you when you have
performed a never-before-done
purchase.
• You do not see less-important mails
(spam) in your inbox.
• Search engines shows ads based on
your previous searches.
• FB adjusts your News Feed based on
your interaction with others.
www.cammanagementsolutions.com

3. DEFINITION
“Field of study that gives computers the ability to
learn without being explicitly programmed”
~ Arthur Samuel – 1959
www.cammanagementsolutions.com

4. DEFINITION
“A computer program is said to learn
from experience E with respect to some
class of tasks T and performance
measure P, if its performance at tasks in
T, as measured by P, improves with
experience E”
~ Tom M. Mitchell
“It is all about making sense from data,
we are asking the computer to make
some sense from data”
www.cammanagementsolutions.com

5. WHY WE NEED THIS?
• Discover the knowledge that is previously unseen from large
datasets.
– Identifying the risk factor on certain operations
• Provide solutions that can be automatically adjusted based on
users’ input (based on individuals behaviors).
– Gaming
• Facilitate to build
hard-to-implement software
solutions for seeing probabilities
– Identifying 3D objects
www.cammanagementsolutions.com

6. KEY TERMS
* Training Set
www.cammanagementsolutions.com
Features/Attributes Target Attribute
Instances/Examples
• Expert system
• Knowledge representation
• Models
• Modules
• Tasks
• Algorithms

7. MACHINE LEARNING – CATEGORIZATIONS I
• Supervised learning
– Instructing the computer to learn the relationships between inputs
and outputs by giving an example inputs and desired output.
• Unsupervised learning
– Instructing the computer to find hidden patterns within the dataset
given without explicitly marking inputs.
• Reinforcement learning
– Computer program interacts with a dynamic environment targeting
a certain goal, without instructing whether it has come close to the
target or not.
www.cammanagementsolutions.com

8. MACHINE LEARNING – CATEGORIZATIONS II
• Classification
– Target values are called as “classes” and assumed to be a finite number of
classes.
– Predicts what class an instance of data should fall into.
• Regression
– Predicts a numeric values, outputs are continuous rather than discrete.
• Clustering
– Group similar items together
• Density estimation
– Shows statistical values that describes data
• Dimensionality reduction
– Distils data down to only the important information and remove the noise.
www.cammanagementsolutions.com

9. ALGORITHMS
www.cammanagementsolutions.com

10. HOW IT WORKS?
Training set Pre-Processing
www.cammanagementsolutions.com
Dimensionality reduction Model Learning
Test set
Trained Model
Model Testing
Input
Output
- Features and target variables
- Purpose
- Credit Amount
- Other debtors
- Age
-
- Feature selection
- Feature projection
- Classification
- Regression
- Clustering
- Density estimation

11. USE CASES
• Google Brain
– Deep Learning project (2011) – Type of AI and ML
– Later named as “Google Brain (2012)
– Able to recognize a “cat” based on 10 million digital images (from
YouTube) using 16,000 computers, mimicking some human brain activities.
– Currently used in;
• Android Operating System’s speech recognition system
• Photosearch for Google+
www.cammanagementsolutions.com

12. USE CASES
• Marketing
– Churn – to identify churners early
– Customer segmentation – grouping customers
for promotion
• Risk
– Credit risk - can credit be granted to the
customer
– Fraud detection - detect invalid/odd
transactions
• Sales
– Forecasting – see the trends of sales for
adjusting processes
www.cammanagementsolutions.com

13. DEMO – PREDICTING CREDIT RISK
www.cammanagementsolutions.com
Training Set

14. DEMO – PREDICTING CREDIT RISK
www.cammanagementsolutions.com
Experiment

15. DEMO – PREDICTING CREDIT RISK
www.cammanagementsolutions.com
Preprocessing

16. DEMO – PREDICTING CREDIT RISK
Testing with multiple
algorithms
www.cammanagementsolutions.com
Evaluating the
model

17. DEMO – PREDICTING CREDIT RISK
Trained model
www.cammanagementsolutions.com

18. DEMO – PREDICTING CREDIT RISK
www.cammanagementsolutions.com
Setting input
Calling the service
and see the output

19. WILL BE DISCUSSING MORE ON THIS………
www.cammanagementsolutions.com

20. www.cammanagementsolutions.com
SLASSCOM TECH TALKS
www.facebook.com/SlasscomTechnologyForum
http://www.slasscom.lk/events
https://twitter.com/slasscom
www.slideshare.net/slasscomtechforum