You are building a model to predict whether or not it will rain on a given day. You have thousands of input features and want to see if you can improve training speed by removing some features while h

Question

Options

• AEliminate features that are highly correlated to the output labels.
• BCombine highly co-dependent features into one representative feature.
• CInstead of feeding in each feature individually, average their values in batches of 3.
• DRemove the features that have null values for more than 50% of the training records.

• A
  4% (1)
• B
  84% (21)
• C
  4% (1)
• D
  8% (2)

Explanation

Option B is correct because combining highly co-dependent (correlated) features - a technique called dimensionality reduction (e.g., PCA) or feature engineering - reduces the total number of inputs without discarding the predictive signal they carry, thereby speeding up training while preserving accuracy.

Why the distractors fail:

• A is backwards - features correlated with the output label are your most valuable predictors; removing them would directly harm accuracy.
• C is arbitrary and statistically unsound; averaging unrelated features together destroys meaningful signal and introduces noise with no principled basis.
• D may remove useful features unnecessarily - missing data can often be imputed, and a feature with 51% nulls might still carry strong predictive value for the rows where it exists.

Memory tip: Think "co-dependent = redundant." If two features are highly correlated with each other (not the label), they're telling the model the same story twice - merging them trims the redundancy. The key contrast to burn in: correlated with the label = keep it; correlated with another feature = candidate for combining.

No community discussion yet for this question.