All MCQs

Binning maps a continuous or finely measured variable into intervals. It may improve interpretability but also discards within-bin detail.

Ordinal categories have a meaningful sequence, so codes can reflect that ranking. The numerical gaps should not automatically be interpreted as equal unless justified.

One-hot encoding creates separate binary indicators without imposing an artificial order. This is appropriate for nominal categories.

Bag-of-words converts text into a vector of term counts or indicators. It generally discards word order while preserving vocabulary usage.

Timestamps contain multiple temporal components that may reveal cycles or seasonal patterns. Extracting those components creates model-ready features.

A squared term is a second-degree polynomial transformation of the original variable. It allows a model to represent certain curved relationships.

An interaction feature represents the joint effect of two variables rather than treating their effects as purely separate. It can capture situations where one predictor’s influence depends on another.

Feature generation transforms raw information into representations that expose useful patterns to a model. Good features can improve performance even when the algorithm remains unchanged.

Features are measurable attributes supplied to an algorithm, such as age, income, or word count. They may be raw measurements or values constructed from raw data.

Priors allow the posterior calculation to reflect that some classes occur more often than others. They should be chosen or estimated carefully to match the intended deployment setting.

Strongly correlated features violate conditional independence and can make the model overstate their combined support. Classification may remain useful, but probability estimates can be distorted.

The conditional-independence factorization keeps estimation and prediction computationally manageable. Sparse count or presence features fit naturally with common Naive Bayes variants.