Learning an interpretable model?

Question

I am working on problems in the field of medical imaging where the need for a simple and interpretable model is important from a clinical perspective. This means that I have to explain the algorithm's prediction to non-experts (well non-experts in Mathematics).

My questions is twofold:

As far as I know, an interpretable model is a model that assigns weights to each feature or weak classifier to form a strong classifier. Are there other methods that make a model interpretable?
Which classification methods are interpretable? I know that the linear SVM and AdaBoost are. So, are there other methods?

score 4 · Accepted Answer · answered Jul 25 '13 at 08:14

4

The most obvious choice is logistic regression, which is by far the most commonly used classifier in medical contexts. Kernelized versions of LR exist too, in case you need something more complex.

Next to that: anything based on decision trees and rule mining would work. I would say linear SVM is a border case.

answered Jul 25 '13 at 08:14

Marc Claesen

17,399
1
49
70

Ah I forgot to mention LR...Still, what is the reason for preferring LR over other methods? In what sense would it be better? – Leeor Jul 25 '13 at 08:23
1

LR directly provides information about the relevance of every predictor. Other methods generally do not. – Marc Claesen Jul 25 '13 at 08:26
Information besides the weights? – Leeor Jul 25 '13 at 08:30
Significance levels for predictors. – Marc Claesen Jul 25 '13 at 08:38

Learning an interpretable model?

1 Answers1