All MCQs

Forward selection evaluates candidate additions to the current subset using a chosen criterion. The process continues until further additions provide insufficient benefit.

Recursive feature elimination uses model-based importance to shrink the candidate feature set step by step. Validation can help determine how many features to retain.

Near-constant variables often provide little ability to distinguish observations. A variance threshold removes them without using the target.

Mutual information measures reduction in uncertainty and is not restricted to linear association. It can reveal dependence patterns that have near-zero Pearson correlation.

A feature can be useful through nonlinear patterns or in combination with other variables despite weak marginal correlation. Correlation-based screening may therefore discard predictive information.

Embedded methods incorporate selection into the learning procedure, such as through coefficient penalties or tree splits. They sit between fast filters and costly wrappers.

Wrapper methods use model performance as the criterion for evaluating candidate subsets. They can capture feature interactions but may be computationally expensive.

Filter methods rank or screen features using properties such as association, variance, or information scores. They operate largely independently of the final predictive algorithm.

Standardization subtracts the feature mean and divides by its standard deviation. The transformed training values then have mean zero and standard deviation one, subject to convention.

Min-max normalization subtracts the minimum and divides by the observed range for a 0-to-1 mapping. It preserves relative spacing but is sensitive to extreme endpoints.

Algorithms based on distance or coefficient penalties can be dominated by high-range variables. Scaling controls the effect of units without necessarily changing ordering.

A log transformation reduces the influence of very large values and can stabilize relative changes. It is appropriate only when the domain and allowed values support it.