Knowledge check

Quiz

In Progress

Knowledge check

Evaluate your understanding of this unit by completing the knowledge check below.

What is the primary purpose of polynomial features and feature crossing in numerical data?

To reduce the number of variables in a datasetTo smooth out skewed distributionsTo capture non-linear relationships between featuresTo standardise variable scales

Which transformation is most appropriate for normalising positively skewed data with only positive values?

Yeo-JohnsonZ-score standardisationBox-Cox transformationMin-max scaling

One-hot encoding is best used when:

The categorical variable is nominal with low cardinality.The variable is ordinal and the order should be preserved.The variable has thousands of unique categories.The data contains continuous numerical values.

What is a major risk when using target encoding without safeguards?

Data duplicationTarget leakage and overfittingReduced feature cardinalityPoor feature importance tracking

TF-IDF improves on Bag-of-Words by:

Capturing word order and grammarEncoding semantic meaning using vectorsWeighting rare but informative words more heavilyIgnoring punctuation and stopwords

Word embeddings like Word2Vec and GloVe are useful because they:

Reduce dimensionality using PCACapture semantic similarity between wordsProvide simple word frequency countsDetect named entities in text

Histogram of Oriented Gradients (HOG) is primarily used to:

Identify dominant colour channelsDetect texture using local patchesDescribe edge orientations for shape recognitionReduce image resolution

What is the primary benefit of dimensionality reduction for image features?

Increasing model accuracy by adding more featuresCompressing high-dimensional data while preserving important structureImproving colour contrastExtracting gradients and edges

A lag feature allows a model to:

Incorporate a previous value of a feature as a predictorSummarise a window of future valuesNormalise time-based inputLabel anomalies in time-series data

What does seasonality extraction aim to capture?

Overall trend in the dataExtreme outliers in residualsRecurring calendar-based patternsMissing time intervals

In progress