Hello,

LDA in effect projects the input data into a space where data samples belonging to the same class are "close" to each other. This means that after the use of LDA, you end up with a projection of the original input data, which you can theoretically use with any classifier of interest. You can look at the sklearn implementation here: https://scikit-learn.org/stable/modules/generated/sklearn.discriminant_analysis.LinearDiscriminantAnalysis.html which defines a transform() method, which will give you the raw LDA scores that can then be classified using any method of your choice.

There is also a deep network-based version of LDA called DeepLDA which uses a deep neural network to perform the projection, and the projected data points are classified using a SVM. You can find the code for that here: https://github.com/VahidooX/DeepLDA/blob/master/code/DeepLDA.py

Hope this helps.

Good luck!

Article Linear discriminant analysis: A detailed tutorial

Hello Mátyás Bukva ,

Of course, you can use the data which is reduced by LDA or any other dimension reduction technique such as LLE, t-sne, etc. You can use any dataset which is reduced by any d-reduction technique as an input of your classifiers such as SVM, or CNN. You can check the work in the link below:

Conference Paper A Comparative Study on Electronic Nose Data Analysis Tools

In this work, firstly we take a dataset and apply several d-reduction methods. Then, we feed our system with these d-reduced datasets to classify the types of gases (which is collected via electronic nose) to analyze them to see which d-reduction technique is well suited for e-nose studies.

I hope it will help you.

Stay safe and healthy.