CODEWITHSUNDEEP
r
Sevvandi Kandanaarachchi
Sevvandi Kandanaarachchi

Reputation: 11

Use Linear Discriminant Analysis for dimension reduction

I want to use LDA for dimensionality reduction. I am using R. The examples that I found used LDA mostly for classification. So how does one use LDA for dimensionality reduction ? Is there an inbuilt function call in R that does dimensionality reduction, or do you have to code it up ?

Thank you. Sevvandi

Upvotes: 1

Views: 1576

Answers (1)

damienfrancois
damienfrancois

Reputation: 59072

LDA is primarily a dimensionality reduction technique, similar to PCA except that is aims at taking into account the class label of the data.

Often, it is used to project onto one dimension, the Fisher linear discriminant which allows determining a threshold above which one class is predicted, and below which the other is. This Fisher linear discriminant is the eigen vector of the product of the inverse of the within class scatter by the between class scatter, corresponding to the largest eigenvalue.

But you can choose as many eigen vector as there are dimensions, you are not limited to just one. I believe the eigenvectors can be found in the scaling output parameter of R's lda function.

See for instance Theodoris (2008) chapters 5.8, 6.1-6.3 for more information.

Upvotes: 1

Related Questions