using unsupervised data for testing in product development experiments conference ניסויים...

16
Unsupervised data the big promise for Testing Edith Ohri Tel. 054-3179161 [email protected]

Upload: edith-ohri

Post on 10-Apr-2017

124 views

Category:

Data & Analytics


0 download

TRANSCRIPT

Unsupervised data the big promise for

Testing

Edith Ohri Tel. 054-3179161

[email protected]

Testing stands for -

Ensuring that a product is functioning well and in compliance with relevant standards, and that it can sustain rough conditions and handling.

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 2

The problem in testing new products

New product functions don’t have yet standards, the testing therefore becomes more complicated expensive and lengthy.

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 3

Checking a new helmet in the old days…

Big Data new promise

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 4

Unsupervised data solution

Free data from early testing and external repositories can help in establishing new standards and functionality.

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 5

Pros and Cons

Unsupervised data are indeed in abundance, and rich in information, which enable shorter and more productive R&D, but such data are not representative, they include redundant variables, unknown interrelations, text variables, inconsistencies and errors, which make them unfit to Statistics, UNLESS ..

(Unless one can observe patterns of behavior in that data and create new perfectly reasonable hypotheses – which is what GT data mining does) 14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 6

Using GT patterns of behavior

GT data mining breaks down the data to similarity-homogeneous groups that enable:

• Focus on particular groups

• Identifying exceptions

• Drilldown to root causes

• Discovery

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 7

Example: testing a new polygraph

The polygraph has 8 channels, 5 are new.

The testing is required to validate that the new device identifies liars effectively.

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 8

Method

Focus & Drilldown

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 9

Groups map – liars are marked with red circle

GT findings: lying patterns and indicators

1. High tension throughout the interview – (skin perspiration) ASR is found to be significant if higher than 1.05

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 10

ASR as a function of question no.

0

0.5

1

1.5

2

2.5

0 20 40 60 80 100 120 140

מס' שאלה

AS

R

Poly. (דוברי אמת)

Poly. (שקרנים )

More GT patterns and indicators

2. Larger difference of blood-pressure between arm and head in the first half of interrogation.

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 11

More GT patterns and indicators

3. Faster reaction in “tough” questions

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 12

More GT patterns and indicators

4. Typical camouflage strategies:

a. A restrained reaction before lying followed by a “relief” signal afterward.

b. Irrelevant answers.

c. Inconsistent reactions.

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 13

Benefits of early test by GT

• Confirmation of product performances

• Gauging new standard’s parameters

• Finding hidden patterns

• Early feedback to designers & engineers

• Stronger case for investors, clients and mgm.

• As results, improving the design, reducing risk, and time to market, and increasing chances for success.

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri Slide 14

About

Edith Ohri, Engr. from the Technion (Israel) and MSc from NY polytech. Main field: data analytics algorithms.

Developer of the "GT data mining" solution. GT as software started in 2002, Singapore. It has been applied in diverse fields and proved effective also for Big Data.

Founder of Datalert (startup) and head of Quality branch in the Industrial & Management Engr. Association.

Current ly: Predictive Analytics for health monitoring by IOT, Finances, and Mining Big Data new standard.

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri 15

Thanks

Don’t be afraid to try new things. Remember, Noah Arc was built by amateurs. Professionals built the Titanic.

(- Source unknown)

Edith Ohri

14 Apr 2016 © Using unsupervised data for Testing with GT data mining. All rights reserved, Edith Ohri 16