Business Knowledge Series course

Data mining is an advanced science that can be difficult to do correctly. This course introduces you to the power and potential of data mining and shows you how to discover useful patterns and trends from data. Valuable practical advice, acquired during years of real-world experience, focuses on how to properly build reliable predictive models and interpret your results with confidence. Examples are drawn from several industries, including credit scoring, fraud detection, biology, investments, and cross-selling.

This course is not hands-on training for SAS Enterprise Miner software, although SAS Enterprise Miner is used by the instructor to illustrate specific modeling techniques and by students for their classroom exercises.

Learn how to

• identify projects with a high probability of success
• translate a business problem into a closely related set of technical tasks
• transform raw data to create higher-order features that reveal information
• build models using powerful nonlinear techniques, such as decision trees and neural networks
• use multiple re-sampling techniques to avoid overfit and predict how well models will perform in actual use
• interpret and validate modeling results
• become aware of the top analytic mistakes and how to avoid them.

Who should attend

Those who have a strong interest in solving a business problem, and who have a technical background, especially familiarity with computer programming and statistics

Formats available	Standard duration
Classroom:	2 days

Before attending this course, you should

• have experience with basic modeling, such as regression
• preferably, have prior exposure to SAS Enterprise Miner, though it is not required.

For comprehensive, hands-on training on SAS Enterprise Miner, attend the Applied Analytics Using SAS Enterprise Miner course.

This course addresses SAS Enterprise Miner software.

Executive Summary

• introduction, executive summary of data mining

Algorithms

• data mining algorithms

Mining Process, Fraud Example, Lift Charts, and Decision Trees

• mining process
• fraud detection
• cumulative response charts
• decision trees

Learning Strategies

• types of learning
• nearest neighbors

Neural Networks, Comparing Algorithms

• perceptrons to neural networks
• polynomial networks
• consensus versus contributory methods
• comparing algorithms
• structural adaptation
• greed

Overfit, Resampling

• overfit
• resampling, such as cross-validation, bootstrap, and target shuffling

Visualization

• assumption checking
• model misspecification, outliers, and leverage points
• grand tour
• projection pursuit
• combining visualization with modeling

Ensembles

• weakness of a single model
• combining models, such as Bayesian Model averaging, bagging, and boosting
• generalization: diverge and merge
• academic example: trees with bags of 5 vs. 8 nodes
• industrial example: credit scoring

Top Ten Data Mining Mistakes