Includes Review of Syllabus OVERVIEW OF THE CLASS

Transcription

1 Includes Review of Syllabus OVERVIEW OF THE CLASS

2 What is this class about? This class will introduce data mining The types of problems that can be addressed The methods that can be used Focus will be balanced between learning how to use the methods and understanding how they work A significant class project is required This class is a key component of the new MS in Data Analytics (MSDA) The spring course Machine Learning will cover some methods not covered in this course or covered only superficially 2

3 Textbooks For many years the Intro to Data Mining book by Tan, Steinbach, and Kumar was used One of the commonly used DM textbooks for CS Not always very clear, but other books are not really any better Data Science for Business is much clearer and well written so it also being used. Provides much more on applications of data mining Provides surprising technical depth, so eventually my be the sole book supplemented with other materials Currently in transition. Some readings may be a bit redundant but good to get two perspectives. 3

4 More on the Use of 2 Textbooks Initially I was put off by the use of Business in the Data Science title since this is a CS course But book is still relatively technical and covers some details of the algorithms Also it is critical for a data analyst/scientist to be able to understand how and why to use certain methods and to be able to articulate this. Intro to Data Mining does a poor job at this 4

5 The Class Website & Syllabus The class webiste is at: Includes the syllabus and is linked to the class schedule The class schedule is under active development since this course is being revised from the last time it was offered. Will keep you up to date about what is current and what may still be modified 5

6 Data Mining Lecture 01: Introduction to Data Mining Much of the material in this presentation is not from either textbook 6

7 Let s Start By Seeing What You Know Quick Quiz Do you know what Data Mining is? Do you know of any examples of Data Mining? 7

8 What is Data Mining? Data Mining has many definitions Non-trivial extraction of implicit, previously unknown and potentially useful information from data Exploration & analysis, by automatic or semi-automatic means, of large quantities of data in order to discover meaningful patterns 8

9 Alternative Names Data Mining was/is known by these other names (although many of these have lost favor over time): Knowledge discovery in databases (KDD) Knowledge extraction Data/pattern analysis Data archeology, data dredging, information harvesting, business intelligence, etc. Recently introduced new names (maybe with different emphases): Data Science Big Data 9

10 What is Big Data? Technically big data means data is so large that conventional data mining methods cannot be applied in normal manner Using this definition, in most cases a data set with 10,000,000 cases is not big data But the term is overused and generally not always interpreted this way Big data technologies include Hadoop Big data technologies used for implementing data mining methods We offer a course in Big Data Programming 10

11 Some Examples Netflix and Amazon use data mining to recommend products (recommender systems) Companies use data mining for marketing Who should be mailed a catalog Who should see what online ads (Google Adwords) Online advertising: large impact Financial companies use credit scoring; fraud detection Customer Churn: who will leave Fordham s WISDM project uses smartphone/watch accelerometer data to classify user activities and perform biometric identification Some search engines cluster retrieved documents into meaningful groups Group pages about Jaguar into car pages and cat pages 11

12 Interesting specific example Wal-Mart used data mining to find out what is needed when a hurricane is coming Strawberry PopTarts increase in sales 7X ahead of a hurricane and the pre-hurricane top selling item is beer. (Data Science for Business page 3) 12

13 A Significant Example Signet bank convinced that modelling profitability, not just default probability, is the way to go But they did not have the proper data Constrained by having data only for strategies they already used Decided to purposefully offer loans in new cases (explore new strategies) Initially poor results but eventually learned from data and go it right Became one of the most successful credit card issuers: Capital One 13

14 Why Data Mining and Why Now? Data Mining was not very popular until about years ago Quick Quiz: What do you think changed? 14

15 Why Mine Data? There are now tremendous amounts of data that are automatically collected and warehoused. What are some examples? Web data, e-commerce Store purchases Bank/Credit Card transactions Cell phone GPS information Smartphone and Smartwatch Sensor Data 15

16 Why Mine Data? What technological changes have helped make data mining so prevalent now? Computers: cheaper and more powerful Smaller mobile devices are exploding in popularity Disk and other storage: greater capacity and cheaper Increased use of on-line resources and Internet We shouldn t discount the advances in algorithms but most data mining algorithms are relatively mature In business, competitive pressure is strong 16

17 Why Mine Data? Often info hidden in data is not evident Analysts may take weeks to discover useful information Much of the data is never analyzed at all There is just too much data to analyze without assistance 17

18 Scientific Need Data collected at enormous speeds remote sensors on satellite telescopes scanning the skies microarrays generating gene expression data scientific simulations Traditional techniques infeasible 18

19 How Big is the Data? Examples of Large Data Sets AT&T s 26TB call detail database (2003) Ebay 6PB, IRS 150TB data warehouse Yahoo has a 2PB DB to analyze behavior of ½ billion web visitors/month (24 billion events/day) Wal-Mart has a 583 TB database (2006) Indexed web contains about 20 Billion pages Sites like Facebook, Flicker & Twitter contain lots of data Google is estimated (in 2011) to have 900,000 servers to handle its data! 19

20 How Much Data is Being Created? 5 Exabytes new data created (2002, UC Berkeley) Humans created/copied 161/281 Exabytes in 06/07 (IDC) 1 Exabyte = stacks of books stretching from Earth to Sun 3 million times the books ever written Not all data stored at once (includes temporary data) In ZB (2800EB) of data will be created/copied Forecast for 2020: 40 ZB, or (57X number of grains of sand on Earth) OK, we get the point already.! Head hurts. 20

21 Why Data Mining? Why Now? According to BabyCenter.com, today one in three children born in the United States already have an online presence (usually in the form of a sonogram) before they are born. That number grows to 92% by the time they are two. In 2012 the average digital birth of children occurs at approximately six months, with a third of all children s photos and information posted online within weeks of their birth. What will it mean to live in a world where our every moment, from birth to death, is digitally chronicled and preserved in vast cloud based databases, forever? During the first day of a baby s life, the amount of data generated by humanity is equivalent to 70 times the information contained in the library of congress. 21

22 Origins of Data Mining Draws ideas from machine learning/ai, pattern recognition, statistics, and database systems* Traditional techniques may be unsuitable due to Enormity of data High dimensionality Heterogeneous & distributed data Statistics Data Mining Artificial Intelligence Machine Learning Pattern Recognition * databases currently have limited impact; data mining is rarely done in a database but rather on flat files Database systems 22

23 Statistics vs. Data Mining Students familiar from statistics are often confused if differences aren t highlighted When compared to Data Mining: Statistics is more theory-based Data mining methods are based on heuristic algorithms Statistics is based firmly on mathematics (e.g., probability) Statistics is more focused on testing hypotheses vs. finding interesting relationships Statistics makes more assumptions about the data 23

24 The Process of Data Mining Data Mining is a process, formerly referred to as a knowledge discovery process. In this process there is a data mining step that applies data mining algorithms to extract knowledge. About 80% of our class will focus on the data mining step but in the real world 80% of the time is spent on the other steps (e.g., prepping data). The process below was articulated by Fayyad in a seminal paper on Data Mining and KDD. There should be a loop since the process is iterative. 24

25 CRISP Data Mining Process 25

26 Second Part of Introduction: Data Mining Tasks 26

27 Top-Level Data Mining Tasks At highest level, data mining tasks can be divided into: Prediction Tasks (supervised learning) Use some variables to predict unknown or future values of other variables Description Tasks (unsupervised learning) Find human-interpretable patterns that describe the data 27

28 Key Data Mining Tasks Overview of the major data mining tasks studied in this course: Prediction Tasks Classification (and class probability estimation) Regression Description Tasks Clustering Association Rule Discovery Also known as co-occurrence grouping or association rule mining 28

29 Data Mining Tasks Continued Data Science for Business includes several more. These are generally not as basic and may be more application oriented Additional data mining tasks Similarity matching: More of a technique and is always used in clustering and sometimes in classification. Can be of interest in its own right. Profiling, Link Prediction, Data Reduction, and Causal Modeling We will certainly cover similarity matching and may cover some of the others 29

30 Classification: Definition Given a collection of records (training set ) Each record contains a set of attributes, one of the attributes is the class, which is to be predicted. Find a model for class attribute as a function of the values of other attributes. Model maps record to a class value Goal: previously unseen records should be assigned a class as accurately as possible. A test set is used to determine accuracy of the model Class Probability Estimation: estimate the probability that an object belongs to a class Can you think of classification tasks? 30

31 10 10 Classification Example Tid Refund Marital Status Taxable Income Cheat Refund Marital Status Taxable Income Cheat 1 Yes Single 125K No No Single 75K? 2 No Married 100K No Yes Married 50K? 3 No Single 70K No No Married 150K? 4 Yes Married 120K No Yes Divorced 90K? 5 No Divorced 95K Yes No Single 40K? 6 No Married 60K No 7 Yes Divorced 220K No No Married 80K? Test Set 8 No Single 85K Yes 9 No Married 75K No 10 No Single 90K Yes Training Set Learn Classifier Model

32 Classification: Application 1 Direct Marketing Goal: Reduce cost of mailing by targeting a set of consumers likely to buy a new cell-phone product. Approach: Use the data for a similar product introduced before. We know which customers decided to buy and which didn t This {buy, don t buy} decision forms the class attribute Collect various demographic, lifestyle, and company-interaction related information about all such customers. Type of business, where they stay, how much they earn, etc. Use this info as input attributes to learn a classifier model Specific Example KDD Cup is a competition associated with top DM conference The KDD CUP 1998 competition was about direct marketing for a charity. Lots of information is provided 32

33 Classification: Application 2 Fraud Detection Goal: Predict fraudulent cases in credit card transactions Approach: Use credit card transactions and info on account-holders as attributes When and what does customer buy, how often pays on time, etc Label past transactions as fraud or fair transactions. This forms the class attribute. Learn a model for the class of the transactions. Use this model to detect fraud by observing credit card transactions on an account. 33

34 Classification: Application 3 Sky Survey Cataloging Goal: To predict class (star or galaxy) of sky objects, especially visually faint ones, based on the telescopic survey images (from Palomar Observatory) images with 23,040 x 23,040 pixels per image. Approach: Segment the image. Measure image attributes (features) - 40 of them per object. Model the class based on these features. Success Story: Could find 16 new high red-shift quasars, some of the farthest objects that are difficult to find! From [Fayyad, et.al.] Advances in Knowledge Discovery and Data Mining,

35 Classifying Galaxies Courtesy: Early Class: Stages of Formation Intermediate Attributes: Image features, Characteristics of light waves received, etc. Late Data Size: 72 million stars, 20 million galaxies Object Catalog: 9 GB Image Database: 150 GB 35

36 Regression Predict a value of a given continuous (numerical) variable based on the values of other variables Greatly studied in statistics Examples: Predicting sales amounts of new product based on advertising expenditure. Predicting wind velocities as a function of temperature, humidity, air pressure, etc. Time series prediction of stock market indices 36

37 Clustering Given a set of data points find clusters so that Data points in same cluster are similar Data points in different clusters are dissimilar You try it on the Simpsons. How can we cluster these 5 data points? 37

38 What is a natural grouping among these objects? 38

39 What is a natural grouping among these objects? Clustering is subjective Simpson's Family School Employees Females Males 39

40 Clustering Application Market Segmentation: Goal: subdivide a market into distinct subsets of similar customers Approach: Collect different attributes of customers based on their geographical and lifestyle related information. Find clusters of similar customers. Measure the clustering quality by observing buying patterns of customers in same cluster vs. those from different clusters. 40

41 Association Rule Discovery Given a set of records each of which contain some number of items from a given collection Produce dependency rules which will predict occurrence of an item based on occurrences of other items. TID Items 1 Bread, Coke, Milk 2 Beer, Bread 3 Beer, Coke, Diaper, Milk 4 Beer, Bread, Diaper, Milk 5 Coke, Diaper, Milk Rules Discovered: {Milk} --> {Coke} {Diaper, Milk} --> {Beer} Diapers beer 41

42 Association Rule Discovery Application Marketing and Sales Promotion Applications When items purchased together one can be used to drive sales of the other Can help determine where to position store items Supermarket shelf management Some stores place bananas in the cereal aisle 42

43 Challenges of Data Mining Scalability Dimensionality Complex and Heterogeneous Data Data Quality Streaming Data Privacy Preservation 43

44 What is (and is not) Data Mining? Based on the definitions of data mining, are these DM or not? Finding a phone number in a directory Not data mining (trivial?, DB query) Grouping related documents returned by search engine Is data mining (not trivial, clustering) Identifying who has a disease based on symptoms Is data mining (not trivial, classification) Web search on keyword using search engine May be data mining** ** More of an information retrieval task than data mining task. However, since Google does much more than keyword matching, there will be a data mining component. For example, Google mines the link structure of the Web to decide which pages are important (link mining is a type of data mining). 44

45 If you are Interested in Data Mining Data sets NYC open data ( UCI Data Repository ( Visit kdnuggets, an online newsletter and more You can arrange to have newsletter ed to you Also includes job openings ACM SIGKDD is the professional organization associated with data mining ACM Special Interest Group (SIG) on data mining Can join SIGKDD for $22 or for $54 can also join ACM as student member 45