Fine tune my SVM in EM

EricTsai — Wed, 01 Jul 2015 16:55:29 GMT

Hi,

I have heard many success story using SVM in predictive modeling but according to my own experience, my SVM didn't outperform my basic Logistic regression in Enterprise Miner.

Could someone help me in fine tuning my SVM? I believe I haven't done a great job in fine tuning my SVM in EMiner and that caused the underperformance of my SVM.

Any papers or any tutorials are more than welcome!

Re: Fine tune my SVM in EM

gergely_batho — Thu, 02 Jul 2015 14:53:23 GMT

I guess you want to compare a linear SVM to your Logistic Regression. (Otherwise it's not fair )

Basically there is only one parameter, that you can use for tuning the linear SVM, it is the penalty value (C).

SVMs do not outperform Logistic Regression in every situation.

They are stable, work very well with high dimensional data...

SVMs are maximal margin classifiers. They work the best if there is a margin. In business terms this means your event and non-event cases are very different. There are only a few "in the middle".

Logistic Regression assumes a smooth transition of the probability as features change.

Don't forget to compare your models on an independent test set.

topic Fine tune my SVM in EM in SAS Data Science

Fine tune my SVM in EM

Re: Fine tune my SVM in EM