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
Chapter 4 An Excel-based Data Mining Tool (iData Analyzer) Jason C. H. Chen, Ph.D. Professor of MIS School of Business Administration Gonzaga University Spokane, WA 99223 [email protected] A/W & Dr. Chen, Data Mining Objectives • This chapter will introduce you the iData Analyzer(iDA) and how to use two of learner models contained in your iDA software of data mining tools. • In Section 4.1 overviews the iDA Model for Knowledge Discovery. • In Section 4.2, introduces an exemplar-based data mining tool, ESX, capable of both supervised learning and unsupervised clustering. • The way of representing datasets and how to use ESX to perform unsupervised clustering and building supervised learning models and others will be also introduced in this chapter. 2 A/W & Dr. Chen, Data Mining 4.1 The iData Analyzer • iDA provides support for the business or technical analyst by offering a visual learning environment, an integrated tool set, and data mining process support. • iDA consists of the following components: – – – – – – Preprocessor Heuristic agent (for larger Large Dataset) ESX Neural Network Rule Maker See p.107 and Appendix A-2 for the instructions of installation Report Generator 3 A/W & Dr. Chen, Data Mining Limitations • The commercial version of iDA is bounded by the size of a single MS Excel spreadsheet, i.e., up to 65,536 rows and 256 columns • The iDA input format uses the first three rows of a spreadsheet to house information about individual attributes – Up to 65,533 data instances in attribute-value format can be mined – The student version allows a maximum of 7,000 data instances (i.e., 7003 rows) After completing the installation if the security setting is high, you should change it to medium and click OK. A/W & Dr. Chen, Data Mining 4 Interface Figure 4.1 – The iDA system architecture Data PreProcessor Large Dataset Yes Heuristic Agent No Mining Technique Neural Networks Explaination No ESX Yes Generate Rules Yes RuleMaker Rules No Report Generator Excel Sheets A/W & Dr. Chen, Data Mining 5 6 A/W & Dr. Chen, Data Mining 4.2 ESX: A Multipurpose Tool for Data Mining • ESX can help create target data, find irregularities in data, perform data mining, and offer insight about the practical value of discovered knowledge. • Features of ESX learner model are: – It supports both supervised learning and unsupervised clustering – It supports an automated method for dealing with missing attribute value – It does not make statistical assumptions about the nature of data to be processed – It can point out inconsistencies and unusual values in data A/W & Dr. Chen, Data Mining 7 Figure 4.3 An ESX concept hierarch Root Level Concept Level Instance Level Root C1 I11 I12 . . . I1j C2 I21 I22 . . . I2k ... Cn In1 In2 . . . Inl 8 A/W & Dr. Chen, Data Mining 4.3 iDAV Format for Data Mining Second Row: C: categorical; R: real-valued Third Row (see Table 4.2 below) Table 4.2 – Values for Attribute Usage Character Usage I The attribute is used as an input attribute U The attribute is not used (categorical attribute with several unique values are of little predictive value) D The attribute is not used for classification or clustering, but attribute value summary information is displayed in all output reports O The attribute is used as an output attribute. For supervised learning with ESX exactly one categorical attribute is selected as the output attribute. 9 A/W & Dr. Chen, Data Mining Table 4.1 – Credit Card Promotion Database: iDAV Format Income Range Magazine Promo Watch Promo Life Ins Promo Credit Card Ins. C C C C C I I I I I 40-50,000 Yes No No No 30-40,000 Yes Yes Yes No 40-50,000 No No No No 30-40,000 Yes Yes Yes Yes 50-60,000 Yes No Yes No 20-30,000 No No No No 30-40,000 Yes No Yes Yes 20-30,000 No Yes No No 30-40,000 Yes No No No 30-40,000 Yes Yes Yes No 40-50,000 No Yes Yes No 20-30,000 No Yes Yes No 50-60,000 Yes Yes Yes No 40-50,000 No Yes No No 20-30,000 No No Yes Yes Sex C I Male Female Male Male Female Female Male Male Male Female Female Male Female Male Female Age R I 45 40 42 43 38 55 35 27 43 41 43 29 39 55 19 10 A/W & Dr. Chen, Data Mining 4.4 A Five-step Approach for Unsupervised Clustering Step 1: Enter the Data to be Mined Step 2: Perform a Data Mining Session Step 3: Read and Interpret Summary Results Step 4: Read and Interpret Individual Class Results Step 5: Visualize Individual Class Rules A/W & Dr. Chen, Data Mining Step 1: Enter The Data To Be Mined 12 A/W & Dr. Chen, Data Mining Step 2: Perform A Data Mining Session 13 A/W & Dr. Chen, Data Mining Figure 4.5 – Unsupervised settings for ESX (#4,p.116) Value for instance similarity: A value closer to 100 encourages the formation of new clusters A value closer to 0 favors new instances to enter existing clusters The real-valued tolerance setting helps determine the similarity criteria for real-valued attributes. A setting of 1.0 is usually appropriate. 14 A/W & Dr. Chen, Data Mining #6 A message box indicating that eight clusters were formed. This tells us the data has been successfully mine. 15 A/W & Dr. Chen, Data Mining #6, #7 (p.116)As a general rule, an unsupervised clustering of more than five or six clusters is likely to be less than optimal. 16 A/W & Dr. Chen, Data Mining #8 and #9, Repeat steps 1-4. For step 5, set the similarity value to 55 17 A/W & Dr. Chen, Data Mining Re-rule feature Covering set rules: RuleMaker will generate a set of best-defining rules for each class. Minimum correctness rule (50-100): if 80, the rules generated must have an error rate less than or equal to 20% Minimum coverage (10-100): if 10, RuleMaker will generate rules that cover 10% or more of the instances in each class. Attribute significance (start with 80-90): values close to 100 will allow RuleMaker to consider only those attribute values most highly predictive of class membership for rule generation. 18 A/W & Dr. Chen, Data Mining #10 (p.117) Set minimum rule coverage at 30 30 Minimum correctness rule (50-100): if 80, the rules generated must have an error rate less than or equal to 20% Minimum coverage (10-100): if 10, RuleMaker will generate rules that cover 10% or more of the instances in each class. Attribute significance: values close to 100 will allow RuleMaker to consider only those attribute values most highly predictive of class membership for rule generation. 19 A/W & Dr. Chen, Data Mining A Production Rule for the Credit Card Promotion Database IF Sex = Female & 19 <=Age <= 43 THEN Life Insurance Promotion = Yes Rule Accuracy: 100.00% Rule Coverage: 66.67% Question: Can we assume that two-thirds of all females in the specified age range will take advantage of the promotion? • Rule accuracy is a between-class measure. • Rule coverage is a within-class measure. A/W & Dr. Chen, Data Mining Output Reports: Unsupervised Clustering • RES SUM: This sheet contains summary statistics about attribute values and offers several heuristics to help us determine the quality of a data mining session. • RES CLS: this sheet has information about the clusters formed as a result of an unsupervised mining session • RUL TYP: Instances are listed by their cluster number. The typicality of instance i is the average similarity of i to the other members of its cluster. • RES RUL: The production rules generated for each cluster are contained in this sheet. 21 A/W & Dr. Chen, Data Mining #10 (p.117) Set minimum rule coverage at 30 22 A/W & Dr. Chen, Data Mining Figure 4.7 Rules for the credit card promotion database 23 A/W & Dr. Chen, Data Mining Step 3: Read and Interpret Summary Results (p.117) (Sheet1 RES SUM) • Class Resemblance Scores (RES) • Domain Resemblance Score • Domain Predictability 24 A/W & Dr. Chen, Data Mining Step 3: Read and Interpret Summary Results (p.119) Instances of Class 1 have a best withinclass fit Similarity value (within the class) In general, the within-class RES scores should be higher than the domain RES. It should be true for most of the 25 classes. A/W & Dr. Chen, Data Mining Figure 4.9 - Step 3: Read and Interpret Summary Results (cont.) 26 A/W & Dr. Chen, Data Mining Figure 4.9 -Statistics for numerical attributes and common categorical attribute values Step 3: Read and Interpret Summary Results (cont.) 27 A/W & Dr. Chen, Data Mining Step 4: Read and Interpret Individual Class Results (p.121) (Sheet1 RES CLS) • Typicality – is defined as the average similarity of an instance to all other members of its cluster or class • Class Predictability is a within-class measure. – the percent of class instances having a particular value for a categorical attribute • Class Predictiveness is a between-class measure – it is defined as probability an instance resides in a specified class given the instance has the value for the chosen attribute 28 A/W & Dr. Chen, Data Mining Figure 4.10 – Class 3 Summary Results within-class between-class 29 A/W & Dr. Chen, Data Mining Figure 4.11 – Necessary and sufficient attribute values for Class 3 30 A/W & Dr. Chen, Data Mining Step 5: Visualize Individual Class Rules IF life ins Promo = Yes THEN Class = 3 :rule accuracy 77.78% :rule coverage 100.00% 31 A/W & Dr. Chen, Data Mining 4.5 A Six-Step Approach for Supervised Learning • Step 1: Choose an Output Attribute – Launch a fresh life insurance promotion • Step 2: Perform the Mining Session • Step 3: Read and Interpret Summary Results • Step 4: Read and Interpret Test Set Results • Step 5: Read and Interpret Class Results • Step 6: Visualize and Interpret Class Rules 32 A/W & Dr. Chen, Data Mining Step 2: Perform the Mining Session Filename: CreditCardPromotion-supervised.xls O: output; D: Display-Only 33 A/W & Dr. Chen, Data Mining Step 2(#4): Select the number of instances for training and a real-valued tolerance setting (p.127) 34 A/W & Dr. Chen, Data Mining Step 3 – Read and Interpret Summary Results Domain statistics for categorical attributes tells us that 80% of the training instances represent individuals without credit card insurance. 35 A/W & Dr. Chen, Data Mining Step 3 – Read and Interpret Summary Results (cont.) 36 A/W & Dr. Chen, Data Mining Step 4 - Read and Interpret Test Set Results 37 A/W & Dr. Chen, Data Mining Step 5 - Read and Interpret Results for Individual Classes (p.130) 38 A/W & Dr. Chen, Data Mining Sheet1 RUL TYP In Class Yes (Life Ins. Promo) Instances of Credit Card Ins = Yes is 40% (2/5) 39 A/W & Dr. Chen, Data Mining Step 6 – Visualize and Interpret Class Rules (p.130) Re-rule feature Covering set rules: RuleMaker will generate a set of best-defining rules for each class. Minimum correctness rule (50-100): if 80, the rules generated must have an error rate less than or equal to 20% Minimum coverage (10-100): if 10, RuleMaker will generate rules that cover 10% or more of the instances in each class. Attribute significance (start with 80-90): values close to 100 will allow RuleMaker to consider only those attribute values most highly predictive of class membership for rule generation. 40 A/W & Dr. Chen, Data Mining 4.6 Techniques for Generating Rules 1. 2. 3. 4. 5. Define the scope of the rules. Choose the instances. Set the minimum rule correctness. Define the minimum rule coverage. Choose an attribute significance value. A/W & Dr. Chen, Data Mining 4.7 Instance Typicality Typicality Scores • Identify prototypical and outlier instances. • Select a best set of training instances. • Used to compute individual instance classification confidence scores. A/W & Dr. Chen, Data Mining Figure 4.13 Instance Typicality 43 A/W & Dr. Chen, Data Mining 4.8 Special Considerations and Features • Avoid Mining Delays • The Quick Mine Feature – Supervised with more than 2000 training set instances, “quick mine” feature will be asked – Unsupervised with more than 2000 data instances. ESX is given a random selection of 500 instances. • Erroneous and Missing Data 44 A/W & Dr. Chen, Data Mining Homework • Use EXS (and iDA) to perform a supervised data mining session using the CardiologyCategorical.xls data file. • Save output file as CardiologyCategorical-supervised.xls • Lab#4 (p.141) • Turn in – 1. Spreadsheet file (CardiologyCategoricalsupervised.xls) that contains the outcome of data mining session – 2. Word file that includes (and explains) answers to all questions (a. thru n.) 45 A/W & Dr. Chen, Data Mining