1
$\begingroup$

I want to build a classifier from a dataset taken from human tissue with 3 classes: no disease, disease A and disease B. Each group ostensibly has data from 10 patients, and each patient has ~30 observations (measurements taken from different locations within a labelled region).

However the no disease group actually shares 8 patients from disease A - just tissue from a different region. The disease B group patients are independent.

First, I am correct of being suspicious of the dependance in this data?

If so, is there a way to take into account this dependancy whilst still classifying into one of the 3 classes?

There is a further complication that actually groups A and B each have one patient twice, just measurements from different regions (still disease A and B respectively though), though I feel this is a separate issue.

Improve this question
$\endgroup$
0

1 Answer 1

Reset to default
1
$\begingroup$

First, I am correct of being suspicious of the dependance in this data?

yes

If so, is there a way to take into account this dependancy whilst still classifying into one of the 3 classes?

Have a look into one-class classification.

One-class classification models each class independently of all other classes, and conceptually aligns well with many general medical diagnostic applications. In medical diagostics, discriminative classification (the "usual" classifiers) would typically be appropriate for differential diagnostics. The main difference between the situations being that in differential diagnostics you do know that the patient in question can only belong into one of a few pre-known categories - whereas in general, a patient may have a number of known or even unknown conditions or diseases.

Improve this answer
$\endgroup$
2
  • $\begingroup$ So you would consider each class in turn as the target class and build an ensemble of classifiers? $\endgroup$
    – N Blake
    Commented Mar 26, 2021 at 22:44
  • $\begingroup$ An ensemble of classifiers (as in ensemble model/prediction or aggregated model/prediction) would usually refer to a bunch of models doing the same prediction. But in the every-day meaning of "ensemble", yes, one-class classification would mean building a set or ensemble of models that each predict membership for a different class. $\endgroup$
    – cbeleites
    Commented Mar 27, 2021 at 19:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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