LDA is most commonly used to solve classification issues with a categorical output variable. It can be used for both binary and multi-class categorization. The Gaussian Distribution of the input variables is used in the conventional LDA model.
Before the classification procedure, linear discriminant analysis (LDA) is performed to reduce the number of features to a more manageable quantity. Each of the new dimensions created is a template made up of a linear combination of pixel values.
The goal of LDA is to use a linear discriminant function to maximise between-class variance and reduce within-class variance under the assumption that data in each class is characterised by a Gaussian probability density function with the same covariance.
One of the most widely used supervised subspace learning methods is linear discriminant analysis (LDA). When presented with a no-label circumstance, however, LDA will be helpless.
When all of the independent/predictor variables are continuous (rather than categorical) and have a Normal distribution, LDA is effective. However, this is not the case in Logistic Regression, and categorical variables can be employed as independent variables when making predictions.