How to perform linear discriminant analysis (LDA)?

I have a set data (100, 61), 100 is sampling point(observation), 61 is features. I have done data reducement using PCA before, but the plot of coeff 1 and 2 (PCA 1 and 2) has many overlap data. Hence I want to try LDA which has an euclidean distance between classes that expected can perform better than PCA. But I have difficulties to perform LDA using fisheriris function, I refer to this link

https://www.mathworks.com/help/stats/create-and-visualize-discriminant-analysis-classifier.html

but those only plot two feature (PL and PW). But in my case I need to plot 61 features in the same plot as perform in PCA as coeff1/coeff2. How do I can perform data projection using LDA(which considering euclidean distance) ?

Answer 1

[Aditya]

Hi Joynjo,

To perform Linear Discriminant Analysis (LDA) for dimensionality reduction and visualize the results, you can use MATLAB's fitcdiscr function. LDA is particularly useful for maximizing the separation between multiple classes. Here's how you can project your high-dimensional data onto a lower-dimensional space using LDA and visualize the results:

% Example data
% X is your data matrix (100 observations, 61 features)
% Y is your class labels vector (100 observations)
% Replace these with your actual data
% X = ...;
% Y = ...;

% Fit LDA model
ldaModel = fitcdiscr(X, Y);

% Transform data using LDA
[~, score] = predict(ldaModel, X);

% Plot the first two components
figure;
gscatter(score(:,1), score(:,2), Y);
xlabel('LDA 1');
ylabel('LDA 2');
title('LDA Projection');
legend('show');
grid on;

Here is a documentation link for the same:

https://www.mathworks.com/help/stats/discriminant-analysis.html