Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Datamining in e-Business: Veni, Vidi, Vici! Prof. Dr. Veljko Milutinovic: •Coarchitect of the World's first 200MHz RISC microprocessor, for DARPA, about a decade before Intel. •Responsible for several successful datamining-oriented e-business on the Internet products, developed in cooperation with leading industry in the USA and Europe. •Consulted for a number of high-tech companies (TechnologyConnect, BioPop, IBM, AT&T, NCR, RCA, Honeywell, Fairchild, etc...) •Ph.D. from Belgrade. After that, for about a decade, on various positions (professor) at one of the top 5 (out of about 2000) US universities in computer engineering (Purdue). •Author and coauthor of about 50 IEEE journal papers (plus many more in other journals). According to some, a European record for his research field. •Guest editor for a number of special issues of: Proceedings of the IEEE, IEEE Transactions on Computers, IEEE Concurrency, IEEE Computer, etc… •Over 20 books published by the leading USA publishers (Wiley, Prentice-Hall, North-Holland, Kluwer, IEEE CS Press, etc...). •Forewords for 7 of his books written by 7 different Nobel Laureates, in cooperation with Telecom Italia learning services (you are welcome to visit http://www.ssgrr.it). [email protected] http://galeb.etf.bg.ac.yu/~vm Page Number: 1 Issues in Data Mining Infrastructure Authors: Nemanja Jovanovic, [email protected] Valentina Milenkovic, [email protected] Prof. Dr. Veljko Milutinovic, [email protected] http://galeb.etf.bg.ac.yu/~vm Page Number: 2 Data Mining in the Nutshell Uncovering the hidden knowledge Huge n-p complete search space Multidimensional interface NOTICE: All trademarks and service marks mentioned in this document are marks of their respective owners. Furthermore CRISP-DM consortium (NCR Systems Engineering Copenhagen (USA and Denmark), DaimlerChrysler AG (Germany), SPSS Inc. (USA) and OHRA Verzekeringen en Bank Groep B.V (The Netherlands)) permitted presentation of their process model. Page Number: 3 A Problem … You are a marketing manager for a cellular phone company Problem: Churn is too high Turnover (after contract expires) is 40% Customers receive free phone (cost 125$) You pay a sales commission of 250$ per contract Giving a new telephone to everyone whose contract is expiring is expensive Bringing back a customer after quitting is both difficult and expensive Page Number: 4 … A Solution Three months before a contract expires, predict which customers will leave If you want to keep a customer that is predicted to churn, offer them a new phone The ones that are not predicted to churn need no attention If you don’t want to keep the customer, do nothing How can you predict future behavior? Tarot Cards? Magic Ball? Data Mining? Page Number: 5 Still Skeptical? Page Number: 6 The Definition The automated extraction of predictive information from (large) databases Automated Extraction Predictive Databases Page Number: 7 History of Data Mining Page Number: 8 Repetition in Solar Activity 1613 – Galileo Galilei 1859 – Heinrich Schwabe Page Number: 9 The Return of the Halley Comet Edmund Halley (1656 - 1742) 1531 1607 1682 239 BC 1910 1986 2061 ??? Page Number: 10 Data Mining is Not Data warehousing Ad-hoc query/reporting Online Analytical Processing (OLAP) Data visualization Page Number: 11 Data Mining is Automated extraction of predictive information from various data sources Powerful technology with great potential to help users focus on the most important information stored in data warehouses or streamed through communication lines Page Number: 12 Data Mining can Answer question that were too time consuming to resolve in the past Predict future trends and behaviors, allowing us to make proactive, knowledge driven decision Page Number: 13 Focus of this Presentation Data Mining problem types Data Mining models and algorithms Efficient Data Mining Available software Page Number: 14 Data Mining Problem Types Page Number: 15 Data Mining Problem Types 6 types Often a combination solves the problem Page Number: 16 Data Description and Summarization Aims at concise description of data characteristics Lower end of scale of problem types Provides the user an overview of the data structure Typically a sub goal Page Number: 17 Segmentation Separates the data into interesting and meaningful subgroups or classes Manual or (semi)automatic A problem for itself or just a step in solving a problem Page Number: 18 Classification Assumption: existence of objects with characteristics that belong to different classes Building classification models which assign correct labels in advance Exists in wide range of various applications Segmentation can provide labels or restrict data sets Page Number: 19 Concept Description Understandable description of concepts or classes Close connection to both segmentation and classification Similarity and differences to classification Page Number: 20 Prediction (Regression) Finds the numerical value of the target attribute for unseen objects Similar to classification - difference: discrete becomes continuous Page Number: 21 Dependency Analysis Finding the model that describes significant dependences between data items or events Prediction of value of a data item Special case: associations Page Number: 22 Data Mining Models Page Number: 23 Neural Networks Characterizes processed data with single numeric value Efficient modeling of large and complex problems Based on biological structures - Neurons Network consists of neurons grouped into layers Page Number: 24 Neuron Functionality I1 W1 I2 W2 I3 W3 In f Output Wn Output = f (W1*I1, W2*I1, …, Wn*In) Page Number: 25 Training Neural Networks Page Number: 26 Neural Networks Once trained, Neural Networks can efficiently estimate the value of an output variable for given input Neurons and network topology are essentials Usually used for prediction or regression problem types Difficult to understand Data pre-processing often required Page Number: 27 Decision Trees A way of representing a series of rules that lead to a class or value Iterative splitting of data into discrete groups maximizing distance between them at each split Classification trees and regression trees Univariate splits and multivariate splits Unlimited growth and stopping rules CHAID, CHART, Quest, C5.0 Page Number: 28 Decision Trees Balance>10 Balance<=10 Age<=32 Married=NO Age>32 Married=YES Page Number: 29 Decision Trees Page Number: 30 Rule Induction Method of deriving a set of rules to classify cases Creates independent rules that are unlikely to form a tree Rules may not cover all possible situations Rules may sometimes conflict in a prediction Page Number: 31 Rule Induction If balance>100.000 then confidence=HIGH & weight=1.7 If balance>25.000 and status=married then confidence=HIGH & weight=2.3 If balance<40.000 then confidence=LOW & weight=1.9 Page Number: 32 K-nearest Neighbor and Memory-Based Reasoning (MBR) Usage of knowledge of previously solved similar problems in solving the new problem Assigning the class to the group where most of the k-”neighbors” belong First step – finding the suitable measure for distance between attributes in the data + Easy handling of non-standard data types - Huge models Page Number: 33 K-nearest Neighbor and Memory-Based Reasoning (MBR) Page Number: 34 Data Mining Models and Algorithms Many other available models and algorithms Logistic regression Discriminant analysis Generalized Adaptive Models (GAM) Genetic algorithms Etc… Many application specific variations of known models Final implementation usually involves several techniques Selection of solution that match best results Page Number: 35 Efficient Data Mining Page Number: 36 NO YES Is It Working? Don’t Mess With It! YES Did You Mess With It? You Shouldn’t Have! NO Anyone Else Knows? NO Hide It YES You’re in TROUBLE! NO Can You Blame Someone Else? YES NO PROBLEM! Page Number: 37 YES Will it Explode In Your Hands? NO Look The Other Way DM Process Model 5A – used by SPSS Clementine (Assess, Access, Analyze, Act and Automate) SEMMA – used by SAS Enterprise Miner (Sample, Explore, Modify, Model and Assess) CRISP – tends to become a standard Page Number: 38 CRISP - DM CRoss-Industry Standard for DM Conceived in 1996 by three companies: Page Number: 39 CRISP – DM methodology Four level breakdown of the CRISP-DM methodology: Phases Generic Tasks Specialized Tasks Process Instances Page Number: 40 Mapping generic models to specialized models Analyze the specific context Remove any details not applicable to the context Add any details specific to the context Specialize generic context according to concrete characteristic of the context Possibly rename generic contents to provide more explicit meanings Page Number: 41 Generalized and Specialized Cooking Preparing food on your own Raw Find out what youvegetables? want to eat stake with Find the recipe for that meal Check the Cookbook or call mom Gather the ingredients Defrost the meat (if you had it in the fridge) Prepare the meal Buy missing ingredients Enjoy yourthe food or borrow from the neighbors Clean up everything (or leave it for later) Cook the vegetables and fry the meat Enjoy your food or even more You were cooking so convince someone else to do the dishes Page Number: 42 CRISP – DM model Business understanding Data understanding Data preparation Modeling Business understanding Deployment Evaluation Deployment Evaluation Page Number: 43 Data understanding Data preparation Modeling Business Understanding Determine business objectives Assess situation Determine data mining goals Produce project plan Page Number: 44 Data Understanding Collect initial data Describe data Explore data Verify data quality Page Number: 45 Data Preparation Select data Clean data Construct data Integrate data Format data Page Number: 46 Modeling Select modeling technique Generate test design Build model Assess model Page Number: 47 Evaluation results = models + findings Evaluate results Review process Determine next steps Page Number: 48 Deployment Plan deployment Plan monitoring and maintenance Produce final report Review project Page Number: 49 At Last… Page Number: 50 WWW.NBA.COM Page Number: 51 Se7en Page Number: 52 CD – ROM Page Number: 53 Evolution of Data Mining Evolutionary Step Business Question Enabling Technologies Product Providers Characteristics Data Collection (1960s) What was my average total revenue over the last 5 years? Computers, tapes, disks IBM, CDC Retrospective, static data delivery Data Access (1980s) What were unit sales in New England last March? RDBMS, SQL, ODBC Oracle, Sybase Informix, IBM, Microsoft Retrospective, dynamic data delivery at record level Data Navigation (1990s) What were unit sales in New England last March? Drill down to Boston. OLAP, Multidimensional databases, data warehouses Pilot, IRI, Arbor, Redbrick, Evolutionary Technologies Retrospective, dynamic data delivery at multiple levels Data Mining (2000) What’s likely to happen to Boston unit sales next month? Why? Advanced algorithms, multiprocessors, massive databases Lockheed, IBM, SGI, numerous startups Prospective, proactive information delivery Page Number: 54 Examples of DM projects to stimulate your imagination Here are six examples of how data mining is helping corporations to operate more efficiently and profitably in today's business environment – Targeting a set of consumers who are most likely to respond to a direct mail campaign – Predicting the probability of default for consumer loan applications – Reducing fabrication flaws in VLSI chips – Predicting audience share for television programs – Predicting the probability that a cancer patient will respond to radiation therapy – Predicting the probability that an offshore oil well is actually going to produce oil Page Number: 55 Comparison of fourteen DM tools Evaluated by four undergraduates inexperienced at data mining, a relatively experienced graduate student, and a professional data mining consultant Run under the MS Windows 95, MS Windows NT, Macintosh System 7.5 Use one of the four technologies: Decision Trees, Rule Inductions, Neural, or Polynomial Networks Solve two binary classification problems: multi-class classification and noiseless estimation problem Price from 75$ to 25.000$ Page Number: 56 Comparison of fourteen DM tools The Decision Tree products were - CART - Scenario - See5 - S-Plus The Rule Induction tools were - WizWhy - DataMind - DMSK Neural Networks were built from three programs - NeuroShell2 - PcOLPARS - PRW The Polynomial Network tools were - ModelQuest Expert - Gnosis - a module of NeuroShell2 - KnowledgeMiner Page Number: 57 Criteria for evaluating DM tools A list of 20 criteria for evaluating DM tools, put into 4 categories: Capability measures what a desktop tool can do, and how well it does it - Handles missing data - Considers misclassification costs - Allows data transformations - Includes quality of tesing options - Has a programming language - Provides useful output reports - Provides visualisation Page Number: 58 - Visualisation + excellent capability good capability - some capability “blank” no capability Page Number: 59 Criteria for evaluating DM tools Learnability/Usability shows how easy a tool is to learn and use - Tutorials Wizards Easy to learn User’s manual Online help Interface Page Number: 60 Criteria for evaluating DM tools Interoperability shows a tool’s ability to interface with other computer applications - Importing data - Exporting data - Links to other applications Flexibility - Model adjustment flexibility - Customizable work enviroment - Ability to write or change code Page Number: 61 Data Input & Output Model + excellent capability good capability - some capability “blank” no capability Page Number: 62 A classification of data sets Pima Indians Diabetes data set – – Wisconsin Breast Cancer data set – – 699 instances of breast tumors some of which are malignant, most of which are benign 10 attributes plus the binary malignancy variable per case The Forensic Glass Identification data set – – 768 cases of Native American women from the Pima tribe some of whom are diabetic, most of whom are not 8 attributes plus the binary class variable for diabetes per instance 214 instances of glass collected during crime investigations 10 attributes plus the multi-class output variable per instance Moon Cannon data set – – 300 solutions to the equation: x = 2v 2 sin(g)cos(g)/g the data were generated without adding noise Page Number: 63 Evaluation of forteen DM tools Page Number: 64 Strenghts and Weaknesses Strengths: Weaknesses: Ease of use (Scenario, WizWhy..) Data visualisation (S-plus,MineSet...) Depth of algorithms (tree options) (CART,See5,S-plus..) Multiple nn architectures (NeuroShell) Difficult file I/O (OLPARS,CART) Limited visualisation (PRW,See5,WizWhy) Narrow analyses path (Scenario) Page Number: 65 How to improve existing DM applications The top ten points: Database integration – no more flat files – use the millions $ spent on data warehousing Automated model scoring – without scoring DM is pretty useless – should be integrated with the driving applications Exporting models to other applications – close the loop between DM and applications that need to use the results (scores) Page Number: 66 How to improve existing DM applications Business templates – cross-selling specific application is more valuable than a general modeling tool Effort knob – it is relevant in a way that tuning parametars are not Incorporate financial information – the financial information is very important and often available and should be provided as input to the DM application Page Number: 67 How to improve existing DM applications Computed target columns – allow the user to interactively create a new target variable Time-series data – a year’s worth of monthly balance information is qualitatively different than twelve distinct non-time-series variables Use versus View – do not present visually to user the full model, only the most important levels Wizards – not necessarily but desirable – prevent human error by keeping the user on track Page Number: 68 Potential Applications Data mining has many varied fields of application, some of which are listed below: Retail/Marketing Identify buying patterns from customers Find associations among customer demographic characteristics Predict response to mailing campaigns Market basket analysis Page Number: 69 Potential Applications • Banking Detect patterns of fraudulent credit card use Identify `loyal' customers Determine credit card spending by customer groups Find hidden correlations between different financial indicators Identify stock trading rules from historical market data Page Number: 70 Potential Applications • Insurance and Health Care Claims analysis - i.e., which medical procedures are claimed together Predict which customers will buy new policies Identify behaviour patterns of risky customers Identify fraudulent behaviour Page Number: 71 Potential Applications • Transportation Determine the distribution schedules among outlets Analyse loading patterns • Medicine Characterise patient behaviour to predict office visits Identify successful medical therapies for different illnesses To predict the effectiveness of surgical procedures or medical tests Page Number: 72 Potential Applications • Sport To make the best choice about players in different circumstance To predict the results of relevance match Do a better list of seed players in groups or tournament DM report from an NBA game When Price was Point-Guard, J.Williams missed 0% (0) of his jump field-goal attempts, and made 100% (4) of his jump field-goal-attempts. The total number of such field-goal-attempts was 4. Page Number: 73 DM and Customer Relationship Management CRM is a process that manages the interactions between a company and its customers Users of CRM software applications are database marketers Goals of database marketers are: identifying market segments, which requires significant data about prospective customers and their buying behaviors build and execute campaigns Tightly integrating the two disciplines presents an opportunity for companies to gain competetive advantage Page Number: 74 DM and Customer Relationship Management How Data Mining helps Database Marketing Scoring The role of Campaign Management Software Increasing the customer lifetime value Combining Data Mining and Campaign Management Page Number: 75 DM and Customer Relationship Management Evaluating the benefits of a Data Mining model Gains chart Profability chart Page Number: 76 Data Mining Examples Bass Brewers “We’ve been brewing beer since 1777. With increased competition comes a demand to make faster/better decisions” Northern Bank “The information is now more accessible, paperless, and timely.” TSB Group Plc “We are using Holos because of its flexibility and its excellent multidimensional database” Page Number: 77 Data Mining Examples Delphic University “Real value is added to data by multidimensional manipulation (being able to easily compare many different views of the avaible information in one report) and by modeling.” Harvard - Holden “Sybase technology has allowed us to develop an information system that will preserve this legacy into the twenty-first century.” J.P.Morgan “The promise of data mining tools like Information Harvester is that they are able to quickly wade through massive amounts of data to identify relationships or trending information that would not have been available without the tool.” Page Number: 78 Securities Brokerage Case Study The following four pages are derived from a copyrighted case study originally created by SmartDrill Data Mining (Marlborough, MA, U.S.A.). Their website is: http://smartdrill.com And the original case study appears in its entirety here: http://smartdrill.com/CHAID.html Page Number: 79 Securities Brokerage Case Study Predictive market segmentation model designed to identify and profile high-value brokerage customer segments as targets for special marketing communications efforts. The dependent variable for this ordinal CHAID model is brokerage account commission dollars during the past 12 months We begin by splitting the client's entire customer file into a modeling sample and a validation sample. (Once the model is built using the modeling sample, we apply it to the validation sample to see how well it works on a sample other than the one on which it was built). Page Number: 80 Securities Brokerage Case Study The resulting CHAID model has 55 segments. However, the results are summarized in the following comb chart, showing the segment indexes (indexes of average dollar value) Page Number: 81 Securities Brokerage Case Study The part of Gains Chart: Average Annual Brokerage Commission Dollars Gains chart provides quantitative detail useful for financial and marketing planning. We have highlighted the top 20% of the file in blue The top 20% of the file is worth an average of about $334 per account, which is nearly three times the average account value for the entire sample. … … … … Page Number: 82 … … … … ... Securities Brokerage Case Study Using the data in the gains chart, we can better plan our communications/promotion budget. In general, the best segments represent customers who are experienced, aggressive, self-directed traders. The other decisions that can help us: We might wish to conduct some market research among customers in under-performing segments, or among under-performing customers in the better segments We can use the segment definitions to help us identify possible issues and question areas to include in the survey Before we try to apply such a model, we perform a validation against a holdout sample, to confirm that it is a good model. Page Number: 83 References Bruha, I., ‘Data Mining, KDD and Knowledge Integration: Methodology and A case Study”, SSGRR 2000 Fayyad, U., Shapiro, P., Smyth, P., Uthurusamy, R., “Advances in Knowledge Discovery and Data Mining”, MIT Press, 1996 Glumour, C., Maddigan, D., Pregibon, D., Smyth, P., “Statistical Themes nad Lessons for Data Mining”, Data Mining And Knowledge Discovery 1, 11-28, 1997 Hecht-Nilsen, R., “Neurocomputing”, Addison-Wesley, 1990 Pyle, D., “Data Preparation for Data Mining”, Morgan Kaufman, 1999 www.thearling.com www.crisp-dm.com www.twocrows.com www.sas.com/products/miner www.spss.com/clementine galeb.etf.bg.ac.yu/~vm Page Number: 84 Potentials of R&D in Cooperation with U. of Belgrade An Overview of Advanced Datamining Projects for High-Tech Computer Industry in the USA and EU VLSI Detection for Internet/Telephony Interfaces Goran Davidović, Miljan Vuletić, Veljko Milutinović, Tom Chen, and Tom Brunett * eT Page Number: 86 USERS... ... Superposition/DETECTION Superposition/DETECTION SPECIALIZED INTERNET REMOTE SITE SERVICE PROVIDER HOME/OFFICE/FACTORY AUTOMATION ON THE INTERNET Page Number: 87 Reconfigurable FPGA for EBI Božidar Radunović, Predrag Knežević, Veljko Milutinović, Steve Casselman, and John Schewel* * Virtual Page Number: 88 USERS ... SPECIALIZED INTERNET SERVICE PROVIDER VCC VCC CUSTOMER SATISFACTION vs CUSTOMER PROFILE Page Number: 89 BioPoP Veljko Milutinovic, Vladimir Jovicic, Milan Simic, Bratislav Milic, Milan Savic, Veljko Jovanovic, Stevo Ilic, Djordje Veljkovic, Stojan Omorac, Nebojsa Uskokovic, and Fred Darnell •isItWorking.com Page Number: 90 Testing the Infrastructure for EBI Phones Faxes Email Web links Servers Routers Software • Statistics • Correlation • Innovation Page Number: 91 CNUCE Integration and Datamining on Ad-Hoc Networks and the Internet Veljko Milutinović, Luca Simoncini, and Enrico Gregory *University of Pisa, Santanna, CNUCE Page Number: 92 GSM DM Ad-Hoc Page Number: 93 Internet Genetic Search with Spatial/Temporal Mutations Jelena Mirković, Dragana Cvetković, and Veljko Milutinović *Comshare Page Number: 94 Drawbacks of INDEX-BASED: Time to index + ranking Advantages of LINKS-BASED: Mission critical applications + customer tuned ranking Well organized markets: Best first search If elements of disorder: G w DB mutations Chaotic markets: G w S/T mutations Provider Page Number: 95 e-Banking on the Internet Miloš Kovačević, Bratislav Milic, Veljko Milutinović, Marco Gori, and Roberto Giorgi *University of Siena Page Number: 96 Bottleneck#1: Searching for Clients and Investments 1472++ *University of Siena + Banco di Monte dei Paschi Page Number: 97 SSGRR Organizing Conferences via the Internet Zoran Horvat, Nataša Kukulj, Vlada Stojanović, Dušan Dingarac, Marjan Mihanović, Miodrag Stefanović, Veljko Milutinović, and Frederic Patricelli *SSGRR, L’Aquila Page Number: 98 http://www.ssgrr.it 2000: Arno Penzias 2001: Bob Richardson 2002: Jerry Friedman 2003: Harry Kroto Page Number: 99 Summary Books with Nobel Laureates: Kenneth Wilson, Ohio (North-Holland) Leon Cooper, Brown (Prentice-Hall) Robert Richardson, Cornell (Kluwer-Academics) Herb Simon (Kluwer-Academics) Jerome Friedman, MIT (IOS Press) Harold Kroto (IOS Press) Arno Penzias (IOS Press) Page Number: 100 http://galeb.etf.bg.ac.yu/~vm/ e-mail: [email protected] Page Number: 101