2
$\begingroup$

I am working with classification problem. I have a dataset with a lot of features. A lot of them can easily determine class. On production, I want to ask the user to provide me with only part of the information.

Which algorithm is able to make pretty accurate predictions from incomplete data?

I was thinking about decisions tree, but I don't want to determine the entry point to my algorithm.

Suppose that my features are binary. I want to give choice to my user which features are most common for him and based on that information make pretty accurate predictions.

$\endgroup$
2
$\begingroup$

If your feature set is really huge and most of the variables are strong features and can determine a class on their own, i would try to go with Logistic Regression using elastic net (probably less on the lasso and more on the ridge side, so you do not throw away features which happen to always co-occur in your dataset but not in reality).

Just be sure to only consider features that appear often enough in order to avoid overfitting. This can easily happen, when the number of examples is smaller than the number of features.

Random Forests could also be worth a try, but to my experience if the number of strong features is really great but they all occur relatively rarely, RFs underperform compared to Logistic Regression.

Btw. the fact that the user only reveals partial information might not be a problem at all as long as the information given can determine the class easily - as you said. I would just give it a try.

|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.