Business Knowledge Series course

Presented by Gordon S. Linoff, co-founder of Data Miners, Inc. and co-author of Data Mining Techniques and Mastering Data Mining

This course introduces a data mining methodology that is a superset to the SAS SEMMA methodology around which SAS Enterprise Miner is organized. The course also introduces a wide range of data mining algorithms and both theoretical knowledge and practical skills. In this class, you work through all the steps of a data mining project, beginning with problem definition and data selection, and continuing through data exploration, data transformation, sampling, portioning, modeling, and assessment.

• use a data mining methodology
• build and use decision trees and neural networks for modeling and scoring
• use survival analysis and create survival curves.

Who should attend

Formats available	Duration
Classroom:	3.0 days

No prior knowledge of statistical or data mining tools is required.

This course addresses SAS Enterprise Miner software.

• what is data mining?
• directed and undirected data mining
• models
• profiling and prediction

• why have a methodology?
• how data miners can inadvertently learn things that are not true
• translating business problems into data mining problems
• the importance of model stability
• finding the right input variables
• sampling to create balanced model sets
• partitioning to create training, validation, and test sets
• data preparation
• model assessment

• developing intuition about data
• data structure
• data types
• data values
• exploring distributions
• summary statistics
• histograms
• using SAS Enterprise Miner for data exploration

• the null hypothesis
• statistical significance
• confidence bounds
• variance and standard deviation
• standardized values
• correlation
• linear regression
• logistic regression
• using SAS Enterprise Miner to build regression models

• decision trees as data exploration and classification tools
• decision trees for modeling and scoring
• decision trees for variable selection
• alternate representations of decision trees
• algorithms used to build decision trees
• splitting criteria
• recognizing instability and overfitting in decision tree models
• capturing interactions between variables
• using SAS Enterprise Miner to build decision trees

• origins of neural networks
• neural networks compared with regression
• algorithms used to train neural networks
• data preparation requirements for neural networks
• picking appropriate inputs for neural networks
• creating neural network models using SAS Enterprise Miner

• similarity and distance
• distance metrics appropriate for different kinds of data
• the role of the training set in memory-based reasoning (MBR)
• combining the votes of several neighbors
• other K-nearest neighbor techniques
• collaborative filtering
• using the SAS Enterprise Miner MBR node

• more on similarity and distance
• the k-means algorithm
• divisive clustering
• agglomerative clustering
• data preparation for clustering
• interpreting clusters
• finding clusters with SAS Enterprise Miner

• origins of survival analysis
• how business data is different from clinical data
• hazards and hazard charts
• retention curves and survival curves
• calculating survival from retention
• calculating hazards empirically
• parametric hazard models
• censoring
• competing risks
• survival-based forecasting
• using SAS code in SAS Enterprise Miner to create survival curves

• market basket analysis
• association rules
• sequential pattern analysis
• using SAS Enterprise Miner to discover associations in retail data

• background on graph theory
• sphere of influence
• using link analysis to generate derived variables
• graph-coloring algorithm
• Kleinberg's algorithm

• optimization techniques and problems (SAS/OR software)
• other algorithms
• linear programming problems
• genetic algorithms