0
$\begingroup$

Assume that I have a single feature dataset X comprising of say a 1000 samples. The feature is an integer that may take up any value upto 2^32. Among these, one particular value alone is considered malicious. The target/lable vector flags out the malicious value as 1 and others as 0. What would be the right approach to train a model that could learn this single value and classify it as 1 and all else as 0?

$\endgroup$
2
$\begingroup$

I don't really see this as a learning problem really, if it's the same malicious value you have to watch out for..

Only one "positive" sample (corresponding to the malicious value) would be enough to find the malicious integer.

Have I misunderstood something in your question?

$\endgroup$
2
  • $\begingroup$ You have understood it correctly. Even one sample will do (but you do not know that before hand). However, it is a learning problem. You do not know before hand what is that particular value. I'm possibly oversimplifying. View it this way: Even if I pass it a sample comprising of several features. Among which only one turns out to be the strongest predictor, Can't that be learned? $\endgroup$
    – Mani
    May 23 '17 at 12:00
  • 1
    $\begingroup$ @Mani It can be "learned", but to my understanding this won't really be a model that learned something. Rather, it learned to ignore all other features, focus only on the important one and if it recognizes the "malicious value" for this feature, classify it as a positive example. What I didn't get is, will two positive samples always have the same "malicious value"? If yes, all you have to do is look for this specific value in this feature. If no, then some learning indeed could be done. $\endgroup$
    – Bogas
    May 23 '17 at 13:46

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.