03-05-2018 02:26 PM
Logistic Regression is usually preferred to Linear Discriminant Analysis for a variety of reasons including...
... Linear Discriminant Analysis prefers interval level multivariate normal inputs with identical within-group covariance matrices
... Linear Discriminant Analysis is sensitive to outliers and prefers groups to have a similar size
... Linear Discrimant Analysis often classifies better than Logistic Regression if all the requirements are met but Logistic Regression is more robust when they are not
Consider that data mining problems typically involve large numbers of both numeric and categorical data, many with missing values, and virtually no chance of even the interval variables being multivariate normal and you see why it was not featured in SAS Enterprise Miner. You can, of course, run any of the SAS/STAT procedures in a SAS Code node but this is just using SAS Enterprise Miner to run Base SAS. Coding can become simpler in a SAS Code Node in SAS Enterprise Miner in some cases since you specify inputs based on macro variables defined for the different types of input, but this still requires you to set-up the metadata about your inputs in a prior node.
Hope this helps!
03-05-2018 02:35 PM
thank you for your response
I am trying to ask my class to compare logistic regression and discriminant analysis so the differences among models as you mentioned in your response can be highlighted.