# Linear discriminant analysis in R: how to choose the most suitable model?

The data set vaso in the robustbase library summarizes the vasoconstriction (or not) of subjects’ fingers along with their breathing volumes and rates.

> head(vaso)
Volume  Rate Y
1   3.70 0.825 1
2   3.50 1.090 1
3   1.25 2.500 1
4   0.75 1.500 1
5   0.80 3.200 1
6   0.70 3.500 1


I want to perform a linear discriminant analysis in R to see how well these distinguish between the two groups. And I consider two cases:

ld <- lda(Y ~ ., data=vaso)

ld1 <- lda(Y ~ log(Volume)+log(Rate), data=vaso)


Please help me understand which model is better? What characteristics to look at?

• Your question is not very clear to me, but maybe it's because I'm not familiar with LDA. I understand you only want to measure the impact of the features on Y, right? I'm not sure it's possible to say which variant is "better" if you don't evaluate anything. Also in the second variant you're testing the log of both features, any particular reason for that? Jun 24, 2020 at 16:13
• @Erwan Yes, I really want to measure the impact of the features on Y, but I want to do it in the most appropriate way. And for this I want to choose the most suitable model. Jun 24, 2020 at 16:21

## 1 Answer

I'm not familiar with LDA, but as far as I know you're not really changing the "model" (i.e. the way to measure impact) between the two versions, what you're changing is the features: in the 2nd version, instead of looking at whether the value of the feature impacts Y, you look at whether the log of the value of the feature impacts Y. The first version is of course the most natural way to look at features, the second is common but usually this is used when we already know that the distribution of the feature (or the relation between the feature and the response variable) makes it relevant.