Issues related to datamining 10-26-13d1805

2015-01-13

++ Issues related to datamining 10-26-13d1805

Met with Dr. George Runger on 10-25-13

How to determine distance to svm boundary?
- (Matlab maybe; weka cannot do this) If using linear kernel this is fairly easy.

Distance is distance of a point to the hyperplane represented by w'*x/||w|| = distance where w' is the transpose of the vector weights from the svm x is the values of the point and ||w|| is the norm of w

How to determine range of values for an attribute associated with a class.
-Note that a tree can output 3 ranges for one variable. Using the svm to do this does not sound trivial. All of the other variables have to be constant and then I have to evaluate the value of that attribute at the svm boundary. It's probably just best to look at the attributes individually and see how they vary for one attribute or another.

How an I find out the association of different attributes?
-He felt like using a clustering algorithm with a heatmap was sufficient.