Download Final Exam - Data Mining Processes | MGS 8040 and more Exams Business Management and Analysis in PDF only on Docsity!
MGS 8040 โ Fall 2008
Final Exam
December 8, 2008
Robert Dodson
Question 1
In the Data Mining class we examined 3 primary areas: Data preparation, Exploration & Forecasting. Data preparation
Question 2
Category (Down)Techniques (Across) -> Cluster Analysis Linear Discriminate Analysis Logistic Regression Goal Group observations into clumps/groups Predict something occurring given known parameters and values Predict likelihood that an event or entity will belong to a particular group Example Provide different loan or financial packages based on attributes of an applicant that match a cluster group Using a few attributes about an applicant determine likelihood they would not fail on a loan Whether or not a women continues to use contraception after a year Assumptions - Data Use a smaller refined more specific subset of variables that is portraying characteristics of groups. No dependent variable. The data has known dependent values that are cleaned and standardized. Account for missing and odd data. Generally only two categories of dependent variables. Y values are 0 or 1. Technique Group observations into clusters. The minimum cluster groups you can have are 2. You can many more than 2 clusters. Analyze the cluster results for distinction. Determine optimal cluster group count. Section independent variables into to 10% or smaller groups. Then create dummy variables based on logical grouping within variables from Xtab jobs. Run regression analysis on dummy variables. Eleiminate unnecessary dummies based on P value correlation. Generate model. If necessary run again and eliminate or add dummies.
- Regression on Ys (1 0r 0) โ odds of Y being = 1. Followed by MLE technique to generate algorithm which gives a score. Output Easier observation of a population by segmentation. KS test results "S" Shaped curve and scores for evaluation. Evaluation Provide a picture/view of your applicants/customers/entr ies. To better help with screening and other business approaches. Very subjective with evaluation compared other statistical techniques. KS test results and logical analysis of how the dummies should layout with their values. If not logical remove and run again. The result values are always between 0 and 1 (% from 0 to 100) presenting a weight/score.
Question 3
Dr. Glenn Myatt presented a fascinating lecture on statistical software to help with
matching compounds in the drug/compound arena. First, he explained the basics of how
drug companies go through an expensive process to create, investigate and submit drugs
for approval. Secondly, he gave a quick explanation on the basics of compound structure
nomenclature. Then he displayed the software and exposed the statistical maneuverings
that go on behind the scenes.
The software pulls in a compound structure and breaks it out into a syntax that the
software can use. Visual layout of lines and objects is not going to work with statistical
software. The software has many functions but one is finding all compounds that could
link up to your compound in review and present information about the linking compound
such as toxicity and other side effects and returns scores on other areas of concern.
Fascinating presentation.
