Data Science Feature Engineering and Selection 4

Question 1

Target leakage in feature engineering occurs when a feature:

Accepted Answer

Contains information not available at prediction time

Answer

Has too few categories

Answer

Is perfectly scaled

Answer

Has missing values

Question 2

To avoid data leakage, feature scaling parameters should be fit on:

Accepted Answer

The training set only, then applied to the test set

Answer

The entire dataset before splitting

Answer

The test set only

Answer

Each fold's validation data

Question 3

Mean imputation of missing values can distort which property of a feature?

Accepted Answer

Its variance, which becomes artificially reduced

Answer

Its column name

Answer

Its data type only

Answer

The number of rows

Question 4

Creating a binary 'is_missing' indicator alongside imputation is useful because:

Accepted Answer

Missingness itself may be predictive

Answer

It removes the need to impute

Answer

It always improves scaling

Answer

It eliminates outliers

Question 5

From a timestamp, which is a typical engineered feature?

Accepted Answer

Day of week or hour of day

Answer

The raw string length

Answer

A random hash

Answer

The file size

Question 6

Cyclical features like month are best encoded using:

Accepted Answer

Sine and cosine transformations

Answer

Plain integer 1-12

Answer

One-hot with 1000 columns

Answer

Log transform

Question 7

Why standardize features before applying PCA?

Accepted Answer

So features with larger scales don't dominate the principal components

Answer

To create categorical variables

Answer

To remove all correlation manually

Answer

To increase the number of features