Please help me on choosing right classifer

Question

Hi all,

I am facing a problem on selecting correct classifier for my data-mining task.

I am labeling webpages using statistical method and label them using a 1-4 scale,1 being the poorest while 4 being the best.

Previously,I used SVM to train the system since I was using a binary(1,0) label then.But now since I switch to this 4-class label,I need to change classifier,because I think the SVM classifier will only work for two-class classification(Please correct me if I am wrong).

So could you please offer some suggestion here on what kind of classifier is most approriate here for my classification purpose.

Thanks in advance for suggestions.

Answer 1

There exist multi-class SVMs. LibSVM has an implementation, as does Weka.

Usually it's better to experiment with several classifiers to find out which one works best on your data. The choice of classifier type and training algorithm is far less important than your choice of feature set. You could try naïve Bayes, multi-class SVM, MaxEnt, voted perceptrons, or whatever your library offers.

Answer 2

You are talking about "ordinal classification". It can be done modified using SVM (as already mentioned, it is also implemented in libSVM), using logistic regression, and even using decision trees, or artificial neural networks.

You can even continuize your labels, perform regression analysis of your choice, and then descretize the output. Most of the methods I have mentioned above do that behind the scenes.

Good luck

Answer 3

You might try to check Andrew NG Lecture on how to choose the ML algorithm that bests suits you, I think is quite enlightening, and it might give you some insight on how to manage your data