I am currently conducting my own benchmarking exercise on a dataset making use of the SSAS logistic regression algorithm.
The issue that I currently face is that I am obtaining a very linear set of results, i.e. the co-efficients of the regression equation are all neatly staggered, when run through two alternate software packages other than SSAS. When I run the same
dataset through the SSAS logistic regression algorithm and extract the results, i.e. I extract the co-efficients from the model, the neat linear set of co-efficients become completely non-sensical and do not follow any trend whatsoever.
I am conducting some preperation of the dataset before it is run into each algorithm i.e. I am discretising the dataset into "classes" of my choosing by conducting additional analysis on the underlying data trends. One of my variables is set to "predict
only" while all of the other variables are set as "discrete" and "input". The reason for the use of "discrete" is due to the fact that I am assigning a class ID to each piece of data depending on the which "class" the particular value falls into. As a
result, I obtain a corresponding co-efficient for each class. e.g. variable = Age. I break this into classes 0-18, 19-35, 36 - 74, >=75, assign a class ID and feed this, along with all
View Complete Post