-1
$\begingroup$

I am comparing different classification and clustering methods for analyzing my game (called Memori) data. It is about diagnosing kids with VSMD.

I have fount a dataset from 1980s with only 2 features and it doesn't have any labels (diagnosed/not diagnosed) on it. I managed to add new features for my own dataset, I have about 25 features now. But I have a problem: since I don't have the labels for the original data, I don't know how to be sure that my game diagnoses well.

Any suggestions on how I can proceed?

$\endgroup$
4
  • $\begingroup$ You can work it out and add the labels yourself. It's manual work. $\endgroup$ – SmallChess Feb 27 '17 at 8:42
  • $\begingroup$ you can try to apply a clustering algorithm for external 80s dataset and then manually attach label to those clusters. At least you won't have to manually review every data sample then. $\endgroup$ – dk14 Feb 27 '17 at 8:46
  • $\begingroup$ I would first do some meaningful visualizations, and see if there is a good way to do clustering. Afterwords you can either add labels based on clustering, or do it manually as suggested by @Student T. $\endgroup$ – jpmuc Feb 27 '17 at 9:35
  • $\begingroup$ I think there's a simple answer and the answer is "no". It sounds to me like any method would be cheating. $\endgroup$ – Peter Flom Jun 19 '18 at 11:23
0
$\begingroup$

Don't rely on old data and "magic" to predict ("diagnose") anything.

Get fresh data.

Much of the success we've been seeing from machine learning is because these companies have collected a massive amount of (fresh) data well-suited for their purpose.

So by any means, get new labeled data.

In particular in medicine, which has a lot of false results due to data reuse.

$\endgroup$

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.