0
$\begingroup$

I am doing disease classification (i.e. a person is classified as normal or abnormal) using naive Bayes and SMO classifiers. I have around 30 attributes. Out of these I need to select the most relevant attributes. I am trying to choose the best attributes using Information Gain.

Is this a better method? Please help me out. Thanks in advance.

$\endgroup$
2
$\begingroup$

As you don't have many features, you might try sequential feature selection which selects features based on maximizing an objective function ( training data classification accuracy in most cases). Here is a link for the method. Also, try googling "sequential feature selection".

| cite | improve this answer | |
$\endgroup$
0
$\begingroup$

Have you assessed the correlations that likely exist among the 30 attributes (without regards to disease classification)? By removing one out of every pair of attributes, you will be able to reduce multicollinearity and the number of attributes in your model. This will allow you to remove variables that add little to no additional information as well as reducing your model to a model containing fewer attributes that predict disease classification just as well as the full model containing 30 attributes.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.