Feature scaling and normalization quiz

Question 1

What is feature scaling in machine learning?

Accepted Answer

Transforming numeric features to a comparable range or distribution

Answer

Deleting rows that contain missing values

Answer

Converting text labels into numbers

Answer

Choosing which model to train

Question 2

Which formula correctly describes Min-Max normalization to the range [0, 1]?

Accepted Answer

x' = (x - min) / (max - min)

Answer

x' = (x - mean) / std

Answer

x' = x / max

Answer

x' = (x - max) / (min - max) only when reversed

Question 3

What does standardization (Z-score scaling) produce for a feature?

Accepted Answer

A feature with mean 0 and standard deviation 1

Answer

A feature strictly bounded between 0 and 1

Answer

A feature with all positive integer values

Answer

A feature with median 0 and range 1

Question 4

Why do distance-based algorithms like k-Nearest Neighbors need feature scaling?

Accepted Answer

Because distances are dominated by large-range features without scaling

Answer

Because k-NN cannot read decimal numbers

Answer

Because scaling removes the need for a distance metric

Answer

Because k-NN only works on standardized text

Question 5

Which of these models is essentially unaffected by monotonic feature scaling?

Accepted Answer

Decision trees / random forests

Answer

k-Nearest Neighbors

Answer

Support Vector Machines with RBF kernel

Answer

Neural networks using gradient descent

Question 6

What is the key difference between normalization (Min-Max) and standardization (Z-score)?

Accepted Answer

Min-Max bounds values to a range; standardization centers to mean 0, std 1

Answer

They are identical methods with different names

Answer

Standardization always outputs values in [0,1]

Answer

Min-Max requires the data to be normally distributed

Question 7

A critical rule when scaling is to fit the scaler only on the training data. Why?

Accepted Answer

To avoid data leakage from the test set into training

Answer

Because the test set has no numeric values

Answer

Because scalers can only process one dataset ever

Answer

Because scaling the test set is mathematically impossible

Question 8

Which scikit-learn transformer performs Z-score standardization?

Accepted Answer

StandardScaler

Answer

MinMaxScaler

Answer

LabelEncoder

Answer

OneHotEncoder

Question 9

Which scaler is most robust to outliers because it uses the median and interquartile range?

Accepted Answer

RobustScaler

Answer

MinMaxScaler

Answer

Normalizer

Answer

StandardScaler

Question 10

What does sklearn's Normalizer do, and how is it different from other scalers?

Accepted Answer

It scales each row (sample) to unit norm, not each column

Answer

It standardizes each column to mean 0

Answer

It clips values to [0,1] per column

Answer

It removes outliers from each feature

Question 11

Apply Min-Max normalization to the value 30 in the dataset [10, 20, 30, 40, 50]. What is the result?

Accepted Answer

0.5

Answer

0.3

Answer

0.6

Answer

1.0

Question 12

A feature has mean 50 and standard deviation 10. What is the standardized value of 70?

Accepted Answer

2.0

Answer

1.5

Answer

0.2

Answer

20

Question 13

How do outliers affect Min-Max normalization?

Accepted Answer

An outlier stretches the range and compresses the other values

Answer

Outliers are automatically removed by Min-Max

Answer

Min-Max is completely immune to outliers

Answer

Outliers make all values become exactly 0.5

Question 14

Why does feature scaling help gradient descent converge faster?

Accepted Answer

It makes the loss contours more symmetric so steps are more direct

Answer

It increases the learning rate automatically

Answer

It removes the need for a loss function

Answer

It guarantees the global minimum is reached in one step

Question 15

Does StandardScaler change the shape (distribution) of a feature?

Accepted Answer

No, it shifts and rescales but keeps the same shape

Answer

Yes, it always makes data perfectly normal

Answer

Yes, it removes skewness completely

Answer

No, it converts data into categories

Question 16

Which transformation is most appropriate to reduce strong right-skew before scaling?

Accepted Answer

Log transformation

Answer

Min-Max scaling only

Answer

One-hot encoding

Answer

Adding a constant to every value

Question 17

In a scikit-learn Pipeline, why is it good practice to include the scaler as a step?

Accepted Answer

It prevents leakage during cross-validation and keeps preprocessing consistent

Answer

It makes the model train without any data

Answer

It removes the need to split train and test

Answer

It automatically deletes outliers

Question 18

What does the 'fit' step of a scaler actually compute?

Accepted Answer

The parameters (e.g., min/max or mean/std) used for transforming

Answer

The predictions of the model

Answer

The accuracy of the classifier

Answer

The number of rows in the dataset only

Question 19

Should you scale one-hot encoded binary (0/1) columns the same way as continuous features?

Accepted Answer

Usually no; they are already on a 0/1 scale

Answer

Yes, always multiply them by 100

Answer

Yes, they must be log-transformed first

Answer

No, you must delete them before scaling

Question 20

After fitting MinMaxScaler on training data, a test value is larger than the training max. What happens?

Accepted Answer

The scaled value goes above 1

Answer

It raises an error and stops

Answer

It is automatically clipped to exactly 1

Answer

It becomes negative

Question 21

What is MaxAbsScaler best suited for?

Accepted Answer

Sparse data, scaling to [-1, 1] while preserving zeros

Answer

Removing outliers from dense data

Answer

Converting categories to numbers

Answer

Forcing a normal distribution

Question 22

Why might unscaled features cause problems in L2-regularized models like Ridge regression?

Accepted Answer

Regularization penalizes coefficients unfairly when scales differ

Answer

Ridge cannot accept any numeric input

Answer

Regularization only works on text features

Answer

Scaling makes regularization unnecessary

Question 23

Which pair of statistics does StandardScaler store after fitting?

Accepted Answer

Mean and standard deviation

Answer

Min and max

Answer

Median and IQR

Answer

Mode and range

Question 24

If a feature is constant (all values equal), what problem arises during Min-Max or Z-score scaling?

Accepted Answer

Division by zero because range or std is 0

Answer

The feature doubles in value

Answer

It becomes a perfect normal distribution

Answer

Nothing changes and values stay the same

Question 25

In image processing, pixel values 0-255 are often divided by 255. What kind of scaling is this?

Accepted Answer

Min-Max scaling to [0, 1]

Answer

Z-score standardization

Answer

Log transformation

Answer

One-hot encoding

Question 26

Is scaling required before applying Principal Component Analysis (PCA)?

Accepted Answer

Yes, otherwise high-variance features dominate the components

Answer

No, PCA already ignores scale

Answer

Only if the data has missing values

Answer

Only for categorical features

Question 27

Which sklearn method applies a previously fitted scaling to new data without recomputing parameters?

Accepted Answer

transform

Answer

fit

Answer

fit_transform

Answer

predict

Question 28

What does 'unit variance' mean after standardization?

Accepted Answer

The feature's variance equals 1

Answer

The feature has exactly one unique value

Answer

All values equal 1

Answer

The variance is removed entirely

Question 29

For Support Vector Machines with an RBF kernel, why is scaling especially important?

Accepted Answer

The RBF kernel uses distances dominated by large-range features

Answer

SVMs cannot read floating-point numbers

Answer

Scaling adds more support vectors automatically

Answer

RBF kernels only work on text

Question 30

You scale a feature with MinMaxScaler. Later you need the original values back. What enables this?

Accepted Answer

inverse_transform using stored parameters

Answer

It is impossible to recover original values

Answer

Refitting on the scaled data

Answer

Multiplying everything by 255

Test yourself