In discriminant analysis, which values are calculated from the training data?

• A. Probabilities
• B. Means
• C. Variances
• D. Distances

Answer: (A)

In discriminant analysis, you start by fitting a probabilistic model for each class using the actual training data. A key quantity you derive directly from that data is the prior probability of each class—the proportion of training samples that belong to that class. These priors reflect how common or likely each class is in the data you’ve observed, and they feed into the decision rule that combines with the class-conditional likelihoods to compute posterior probabilities for new observations.

While means and variances (or covariances) are also estimated from the training data to describe each class’s distribution (and are used in the likelihood part of the model), the aspect most directly tied to the training data and often highlighted as being calculated from it is the class probabilities (the priors). Distances aren’t the standard mechanism used in discriminant analysis.