A company wants to predict the success of advertising campaigns by considering the color scheme of each advertisement. An ML engineer is preparing data for a neural network model. The dataset includes

Question

Options

• AApply label encoding to the color categories. Automatically assign each color a unique integer.
• BImplement padding to ensure that all color feature vectors have the same length.
• CPerform dimensionality reduction on the color categories.
• DOne-hot encode the color categories to transform the color scheme feature into a binary matrix.

• A
  4% (1)
• B
  4% (1)
• D
  91% (21)

Explanation

One-hot encoding (D) is correct because neural networks require numerical input, and color categories have no inherent ordinal relationship - one-hot encoding represents each color as a binary vector (e.g., Red = [1,0,0], Blue = [0,1,0]) without implying any false ordering or magnitude between colors.

Why the distractors are wrong:

• A (Label encoding) assigns integers like 1, 2, 3 to colors, which falsely implies an ordering (e.g., "Red < Blue < Green"), causing the model to learn spurious relationships that don't exist.
• B (Padding) is used to standardize sequence lengths in text or time-series data, not to handle categorical features - it's the wrong tool for this problem.
• C (Dimensionality reduction) techniques like PCA are applied after encoding to compress already-numerical features, not as a primary encoding strategy for raw categorical data.

Memory tip: Think of one-hot encoding as giving each category its own "spotlight" - only one light is on at a time, with no category ranked higher than another. Whenever you see unordered categorical data going into a neural network, one-hot encoding is your default choice.

No community discussion yet for this question.