Using Amazon SageMaker Data Wrangler, a data scientist wants to pick predictor variables that best predict a target. The predictors are correlated with the target and the scientist wants to understand variance in different directions across feature space. Which approach meets these needs?

The correct answer is to use Data Wrangler's multicollinearity tools with principal component analysis (PCA) to transform the features and examine variance directions. PCA is a dimensionality reduction technique that transforms correlated variables into a smaller set of uncorrelated variables called principal components. These components capture the maximum variance in the data, allowing the data scientist to understand the variance in different directions across the feature space and identify the most impactful predictors. Multicollinearity tools help identify highly correlated predictors before applying PCA, which can improve PCA's effectiveness. Using multicollinearity measures with VIF quantifies relationships but doesn't transform features or directly show variance directions. The Data Quality and Insights quick model visualization estimates model quality but doesn't specifically address variance directions or feature transformation for prediction. The Data Quality and Insights report ranks features by predictive power but doesn't inherently transform features or explain variance directions in the feature space.