2
$\begingroup$

Consider a classification problem(lets say 2 classes, 'good' and 'bad), where all features are continuous.

what I need is a range of values for each feature that contributes to 'good' classification. What I thought was simply partitioning the feature values based on good or bad label, problem is all values don't equally contribute for good/bad classification.

So what methods can be applied to find such range for each feature ?

$\endgroup$
2
$\begingroup$

It's impossible in general, simply because a particular value or range for feature A might correspond to class 'good' if feature B has a certain value/range but correspond to class 'bad' otherwise. In other words, the features are inter-dependent so there's no way to be sure that a certain range for a particular feature is always associated with a particular class.

That being said, it's possible to simplify the problem and assume that the features are independent: that's exactly what Naive Bayes classification does. So if you train a NB classifier and look at the estimated probabilities for every feature, you should obtain more or less the information you're looking for.

Another option which takes into account the dependency between variables is to train a simple decision tree model: by looking at the conditions in the tree you should see which combinations of features/ranges lead to which class.

| 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.