Data Science: Concepts and Practice

549 Pages·2018·48.729 MB·English

by  Vijay Kotu

Data Science
Data Science
Concepts and Practice
Second Edition
Vijay Kotu
Bala Deshpande
MorganKaufmannisanimprintofElsevier
50HampshireStreet,5thFloor,Cambridge,MA02139,UnitedStates
Copyrightr2019ElsevierInc.Allrightsreserved.
Nopartofthispublicationmaybereproducedortransmittedinanyformorbyanymeans,electronicormechanical,
includingphotocopying,recording,oranyinformationstorageandretrievalsystem,withoutpermissioninwriting
fromthepublisher.Detailsonhowtoseekpermission,furtherinformationaboutthePublisher’spermissionspolicies
andourarrangementswithorganizationssuchastheCopyrightClearanceCenterandtheCopyrightLicensingAgency,
canbefoundatourwebsite:www.elsevier.com/permissions.
ThisbookandtheindividualcontributionscontainedinitareprotectedundercopyrightbythePublisher(otherthan
asmaybenotedherein).
Notices
Knowledgeandbestpracticeinthisfieldareconstantlychanging.Asnewresearchandexperiencebroadenour
understanding,changesinresearchmethods,professionalpractices,ormedicaltreatmentmaybecomenecessary.
Practitionersandresearchersmustalwaysrelyontheirownexperienceandknowledgeinevaluatingandusingany
information,methods,compounds,orexperimentsdescribedherein.Inusingsuchinformationormethodsthey
shouldbemindfuloftheirownsafetyandthesafetyofothers,includingpartiesforwhomtheyhaveaprofessional
responsibility.
Tothefullestextentofthelaw,neitherthePublishernortheauthors,contributors,oreditors,assumeanyliabilityfor
anyinjuryand/ordamagetopersonsorpropertyasamatterofproductsliability,negligenceorotherwise,orfromany
useoroperationofanymethods,products,instructions,orideascontainedinthematerialherein.
BritishLibraryCataloguing-in-PublicationData
AcataloguerecordforthisbookisavailablefromtheBritishLibrary
LibraryofCongressCataloging-in-PublicationData
AcatalogrecordforthisbookisavailablefromtheLibraryofCongress
ISBN:978-0-12-814761-0
ForInformationonallMorganKaufmannpublications
visitourwebsiteathttps://www.elsevier.com/books-and-journals
Publisher:JonathanSimpson
AcquisitionEditor:GlynJones
EditorialProjectManager:AnaClaudiaAbadGarcia
ProductionProjectManager:SreejithViswanathan
CoverDesigner:GregHarris
TypesetbyMPSLimited,Chennai,India
Dedication
To all themothers in our lives
Foreword
A lot has happened since the first edition of this book was published in
2014. There is hardly a day where there is no news on data science, machine
learning, or artificial intelligence in the media. It is interesting that many of
those news articles have a skeptical, if not an even negative tone. All this
underlines two things: data science and machine learning are finally becom-
ing mainstream. And people know shockingly little about it. Readers of this
book will certainly do better in this regard. It continues to be a valuable
resource to not only educate about how to use data science in practice, but
also how the fundamentalconcepts work.
Data science and machine learning are fast-moving fields which is why this
second edition reflects a lot of the changes in our field. While we used to
talk a lot about “data mining” and “predictive analytics” only a couple of
years ago, we have now settled on the term “data science” for the broader
field. And even more importantly: it is now commonly understood that
machine learning is at the core of many current technological breakthroughs.
These are truly excitingtimesfor all the peopleworking inour fieldthen!
I have seen situations where data science and machine learning had an
incredible impact. But I have also seen situations where this was not the
case. What was the difference? In most cases where organizations fail with
data science and machine learning is, they had used those techniques in
the wrong context. Data science models are not very helpful if you only
have one big decision you need to make. Analytics can still help you in
such cases by giving you easier access to the data you need to make this
decision. Or by presenting the data in a consumable fashion. But at the
end of the day, those single big decisions are often strategic. Building a
machine learning model to help you make this decision is not worth
doing. And often they also do not yield better results than just making the
decision on your own.
xi
xii Foreword
Here is where data science and machine learning can truly help: these
advanced models deliver the most value whenever you need to make lots of
similar decisions quickly.Good examples for this are:
(cid:1) Defining the price of aproductin markets with rapidly changing
demands.
(cid:1) Makingoffersfor cross-selling in an E-Commerce platform.
(cid:1) Approving credit or not.
(cid:1) Detecting customerswith ahighrisk for churn.
(cid:1) Stopping fraudulent transactions.
(cid:1) And many others.
You can see that a human being who would have access to all relevant data
could make those decisions in a matter of seconds or minutes. Only that
they can’t without data science, since they would need to make this type of
decision millionsof times, everyday. Consider sifting through your customer
base of 50 million clients every day to identify those with a high churn risk.
Impossible for any human being. But no problem at all for a machine learn-
ing model.
So, the biggest value of artificial intelligence and machine learning is not to
support us with those big strategic decisions. Machine learning delivers most
value when we operationalize models and automate millions of decisions.
One of the shortest descriptions of this phenomenon comes from Andrew
Ng, who is a well-known researcher in the field of AI. Andrew describes what
AI can do as follows: “If a typical person can do a mental task with less than
one second of thought, we can probably automate it using AI either now or
in the near future.”
I agree with him on this characterization. And I like that Andrew puts the
emphasis on automation and operationalization of those models—because
this is where the biggest value is. The only thing I disagree with is the time
unit he chose. It issafeto already go with aminuteinstead of asecond.
However, the quick pace of changes as well as the ubiquity of data science
also underlines the importance of laying the right foundations. Keep in
mind that machine learning is not completely new. It has been an active
field of research since the 1950s. Some of the algorithms used today have
even been around for more than 200 years now. And the first deep learning
models were developed in the 1960s with the term “deep learning” being
coined in 1984. Those algorithms are well understood now. And under-
standing their basic concepts will help you to pick the right algorithm for
the right task.
Tosupportyouwiththis,someadditionalchapterson deeplearningandrec-
ommendation systems have been added to the book. Another focus area is
Foreword xiii
using text analytics and natural language processing. It became clear in the
past years that the most successful predictive models have been using
unstructured input data in addition to the more traditional tabular formats.
Finally, expansion of Time Series Forecasting should get you started on one
ofthe most widely applied data science techniquesin the business.
More algorithms could mean that there is a risk of increased complexity. But
thanks to the simplicity of the RapidMiner platform and the many practical
examples throughout the book this is not the case here. We continue our
journey towards the democratization of data science and machine learning.
Thisjourneycontinuesuntil datascienceandmachinelearningareasubiqui-
tous as data visualization or Excel. Of course, we cannot magically transform
everybody into a data scientist overnight, but we can give people the tools to
help them on their personal path of development. This book is the only tour
guide youneed on this journey.
Ingo Mierswa
Founder
RapidMiner Inc.
Massachusetts, USA
Preface
Our goal is to introduce you toDataScience.
We will provide you with a survey of the fundamental data science concepts
as well as step-by-step guidance on practical implementations—enough to
getyoustartedon this exciting journey.
WHY DATA SCIENCE?
We have run out of adjectives and superlatives to describe the growth trends
of data. The technology revolution has brought about the need to process,
store, analyze, and comprehend large volumes of diverse data in meaningful
ways.However,thevalueofthestoreddataiszerounlessitisactedupon.Thescale
of data volume and variety places new demands on organizations to quickly
uncover hidden relationships and patterns. This is where data science techni-
ques have proven to be extremely useful. They are increasingly finding their
way into the everyday activities of many business and government functions,
whether in identifying which customers are likely to take their business else-
where,or mapping flupandemic using socialmediasignals.
Data science is a compilation of techniques that extract value from data.
Some of the techniques used in data science have a long history and trace
their roots to applied statistics, machine learning, visualization, logic, and
computer science. Some techniques have just reached the popularity it
deserves. Most emerging technologies go through what is termed the “hype
cycle.” This is a way of contrasting the amount of hyperbole or hype versus
the productivity that is engendered by the emerging technology. The hype
cycle has three main phases: peak of inflated expectation, trough of disillu-
sionment, and plateau of productivity. The third phase refers to the mature
and value-generating phase of any technology. The hype cycle for data sci-
ence indicates that it is in this mature phase. Does this imply that data sci-
ence has stopped growing or has reached a saturation point? Not at all. On
the contrary, this discipline has grown beyond the scope of its initial
xv
xvi Preface
applications in marketing and has advanced to applications in technology,
internet-based fields, health care, government, finance, andmanufacturing.
WHY THIS BOOK?
The objective of this book is two-fold: to help clarify the basic concepts
behind many data science techniques in an easy-to-follow manner; and to
prepare anyone with a basic grasp of mathematics to implement these techni-
ques in their organizations without the need to write any lines of program-
ming code.
Beyond its practical value, we wanted to show you that the data science
learning algorithms are elegant, beautiful, and incredibly effective. You will
never look at data the same way once you learn the concepts of the learning
algorithms.
To make the concepts stick, you will have to build data science models.
While there are many data science tools available to execute algorithms and
develop applications, the approaches to solving a data science problem are
similar amongthesetools. Wewantedtopickafullyfunctional,opensource,
free to use, graphical user interface-based data science tool so readers can fol-
low the concepts and implement the data science algorithms. RapidMiner, a
leading data science platform, fit the bill and, thus, we used it as a compan-
ion tool to implement the data science algorithms introduced in every
chapter.
WHO CAN USE THIS BOOK?
The concepts and implementationsdescribed in this book are gearedtowards
business, analytics, and technical professionals who use data everyday. You,
the reader of the book will get a comprehensive understanding of the differ-
ent data science techniques that can be used for prediction and for discover-
ing patterns, be prepared to select the right technique for a given data
problem, andyouwill be able tocreate ageneral-purpose analyticsprocess.
We have tried to follow a process to describe this body of knowledge. Our
focus has been on introducing about 30 key algorithms that are in wide-
spread use today. We presentthesealgorithms in the framework of:
1. A high-levelpracticaluse case for eachalgorithm.
2. Anexplanation of how the algorithm works inplain language. Many
algorithms have astrong foundation in statistics and/or computer
science. In our descriptions, we have tried tostrike abalance between
being accessibleto awider audience andbeingacademicallyrigorous.