If the dependent variable is categorical, the neural network can be used for

• A. Dimensionality reduction
• B. Classification
• C. Regression
• D. Clustering

Answer: (B)

When the target variable is categorical, the task is classification. Neural networks learn to map input features to discrete categories, producing probabilities for each class (often with a softmax layer for multiclass or a sigmoid for binary) and being trained with a loss like cross-entropy that compares predicted probabilities to the true labels. This makes them well-suited for deciding which category an input belongs to.

Dimensionality reduction and clustering, in contrast, are unsupervised techniques that don’t predict a labeled category from inputs. Regression targets continuous values, not categories, so they fit different problem setups.