1
$\begingroup$

I have 5 broad metagenomic "ecoregion" categories (just think lots of DNA at different nice locations) which become the training targets for their complete (and augmented) metagenomic data. Any standard model works fine notably random forest, naive Bayes, SVM, confusion matrix is okay and ROC fine. These are small data sets of 10E5 to 10E6.

The categories are very broad and most predictions (meta genomic data from other "ecoregions"), will fall between these categories. ML in contrast will 'relocate' the prediction into 1 of 5 categories. So thus if I've a "woodland" category and a "lake" category, a marsh will fall "in between" the trained classification, but ML will call it either a 'wood' or a 'lake'.

How do I attain that in-between classification status via ML?

$\endgroup$
2
  • 1
    $\begingroup$ Classification models give a probability (or distance or some packages call a score). What you do with that probability is up to you. If a prediction has a 40% chance of being 1 category, a 30% chance of being another, a 30% chance of being another, that is up to the decision rules to make something out of it. The decision rule can "relocate" using your term to a specific class or not. $\endgroup$
    – Craig
    Nov 4, 2022 at 12:22
  • $\begingroup$ Thanks for this thats useful. Its here stackoverflow.com/questions/49507066/… $\endgroup$
    – M__
    Nov 4, 2022 at 14:14

1 Answer 1

2
$\begingroup$

The task should be reframed as regression or ordinal classification. Using the ecosystem metaphor, a regression target could be the amount of ground covered in water or an ordinal classification target could be sequenced categories of landscapes.

$\endgroup$
1
  • $\begingroup$ Thanks along with @Criag these very good answers. Both the regression and ordinal classification both look extremely interesting solutions and thanks for the clear explanation. $\endgroup$
    – M__
    Nov 5, 2022 at 3:30

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.