0

We are trying to come up with a prediction model for one our products in our team. We did some feature engineering to extract relevant features.

Now we are trying to see which ML methodology would give us the best result. We started with Logistic Regression and SVM. We expected SVM to perform better than Logistic Regression.

I am new to ML concepts. So need some help from expert forum.

Firstly, is it always true that SVM would perform better than Logistic Regression? If not, can someone please help me understand this mathematically? i.e, How do I debug why SVM is under performing compared to logistic regression?

Also is "Binomial Logistic Regression" different from "Linear Logistic Regression"?

Glen_b
  • 257,508
  • 32
  • 553
  • 939
user417397
  • 1
  • 1
  • Do you want to perform a linear or non-linear classification? Logistic regression is designed for linear problems; SVM can manage non linear problems, but you should first choose a non-linear kernel. For linear problems, logistic regression and SVM show data differently. You can see this post: http://stackoverflow.com/questions/26896262/how-does-support-vector-machine-compare-to-logistic-regression – ahstat May 06 '17 at 05:50
  • If your question is not answered by the answers at the indicated duplicate, please clarify your question to make it clear what you're asking in a way that's obviously distinct from that one. – Glen_b May 06 '17 at 06:11
  • @ahstat - thanks for the response. I am using "radial" kernel currently and I was not really was not sure about which way to go. Honestly I did not know how to check whether my data is linearly separable or not as I have around 9 features. I thought will try out both Logistic regression and SVM and see the results. Please suggest if there is a way to figure out the linear separability – user417397 May 27 '17 at 11:48

0 Answers0