Logistic regression quiz

Question 1

What type of problem is logistic regression primarily used to solve?

Accepted Answer

Classification problems

Answer

Continuous value prediction only

Answer

Clustering unlabeled data

Answer

Dimensionality reduction

Question 2

What mathematical function does logistic regression use to map any real number to a value between 0 and 1?

Accepted Answer

Sigmoid function

Answer

ReLU function

Answer

Linear function

Answer

Polynomial function

Question 3

What is the output of the sigmoid function when its input z equals 0?

Accepted Answer

0.5

Answer

0

Answer

1

Answer

Undefined

Question 4

In binary logistic regression, what is the most commonly used default threshold to convert a predicted probability into a class label?

Accepted Answer

0.5

Answer

0.0

Answer

1.0

Answer

0.25

Question 5

What is the loss function typically used to train a logistic regression model?

Accepted Answer

Binary cross-entropy (log loss)

Answer

Mean squared error

Answer

Hinge loss

Answer

Mean absolute error

Question 6

Why is mean squared error generally NOT used as the loss function for logistic regression?

Accepted Answer

It makes the loss surface non-convex

Answer

It is impossible to compute

Answer

It always gives zero gradient

Answer

It requires labeled data

Question 7

What is the 'decision boundary' in logistic regression?

Accepted Answer

The line/surface where predicted probability is 0.5

Answer

The point of maximum loss

Answer

The mean of all data points

Answer

The set of misclassified points

Question 8

Is the decision boundary of standard logistic regression linear or non-linear in the input features?

Accepted Answer

Linear

Answer

Always circular

Answer

Always non-linear

Answer

Random

Question 9

What are the 'log-odds' (logit) in logistic regression?

Accepted Answer

log(p / (1-p))

Answer

p / (1-p)

Answer

log(p)

Answer

1 / (1+p)

Question 10

How do you compute the odds from a probability p?

Accepted Answer

p / (1 - p)

Answer

(1 - p) / p

Answer

p * (1 - p)

Answer

p + (1 - p)

Question 11

What does a positive coefficient (weight) for a feature indicate in logistic regression?

Accepted Answer

The probability of the positive class increases with the feature

Answer

The feature is irrelevant

Answer

The probability decreases with the feature

Answer

The feature must be removed

Question 12

What optimization algorithm is most commonly used to find the best weights in logistic regression?

Accepted Answer

Gradient descent

Answer

K-means

Answer

Apriori

Answer

Backpropagation only

Question 13

In scikit-learn, which class is used to create a logistic regression model?

Accepted Answer

LogisticRegression

Answer

LinearRegression

Answer

SGDClassifier only

Answer

KNeighborsClassifier

Question 14

What does the predict_proba method return in scikit-learn's LogisticRegression?

Accepted Answer

Probabilities for each class

Answer

Only the predicted label

Answer

The loss value

Answer

The model coefficients

Question 15

What is regularization used for in logistic regression?

Accepted Answer

To prevent overfitting by penalizing large weights

Answer

To speed up data loading

Answer

To increase the number of features

Answer

To convert outputs to integers

Question 16

In scikit-learn's LogisticRegression, what does a smaller value of the parameter C imply?

Accepted Answer

Stronger regularization

Answer

Weaker regularization

Answer

No regularization

Answer

More features

Question 17

How can logistic regression be extended to handle more than two classes?

Accepted Answer

Softmax/multinomial or one-vs-rest

Answer

It cannot be extended

Answer

By adding more thresholds only

Answer

By using mean squared error

Question 18

What function generalizes the sigmoid for multiclass logistic regression?

Accepted Answer

Softmax

Answer

ReLU

Answer

Tanh

Answer

Step function

Question 19

Why is feature scaling often recommended before training logistic regression?

Accepted Answer

It speeds convergence and balances regularization

Answer

It changes the class labels

Answer

It removes the need for a threshold

Answer

It is never needed

Question 20

What does the bias (intercept) term represent in logistic regression?

Accepted Answer

The log-odds when all features are zero

Answer

The learning rate

Answer

The number of classes

Answer

The total loss

Question 21

If a logistic regression model outputs a probability of 0.92 for the positive class, what is the predicted label using a 0.5 threshold?

Accepted Answer

Positive class (1)

Answer

Negative class (0)

Answer

Undetermined

Answer

Both classes

Question 22

What is the range of values that the log-odds (logit) can take in logistic regression?

Accepted Answer

Negative infinity to positive infinity

Answer

0 to 1

Answer

-1 to 1

Answer

0 to infinity

Question 23

Which evaluation metric is generally more appropriate than accuracy for a logistic regression model on a highly imbalanced dataset?

Accepted Answer

F1-score or AUC-ROC

Answer

Accuracy only

Answer

Mean squared error

Answer

R-squared

Question 24

What is maximum likelihood estimation (MLE) in the context of logistic regression?

Accepted Answer

Choosing parameters that make observed data most probable

Answer

Minimizing the number of features

Answer

Maximizing the learning rate

Answer

Picking random weights

Question 25

What is the difference between linear regression and logistic regression in terms of their output?

Accepted Answer

Logistic outputs a probability (0-1); linear outputs continuous values

Answer

Both output probabilities

Answer

Both output continuous values

Answer

Logistic outputs only integers from linear values

Question 26

What happens to the sigmoid output as the input z becomes very large and positive?

Accepted Answer

It approaches 1

Answer

It approaches 0

Answer

It approaches 0.5

Answer

It approaches infinity

Question 27

Why does log loss heavily penalize a confident but incorrect prediction?

Accepted Answer

Because -log of a tiny probability becomes very large

Answer

Because it adds the probabilities

Answer

Because it ignores correct predictions

Answer

Because it uses squared error

Question 28

Using the odds interpretation, if a feature's coefficient is b, by what factor do the odds change for a one-unit increase in that feature?

Accepted Answer

By a factor of e^b

Answer

By a factor of b

Answer

By a factor of b squared

Answer

By a factor of log(b)

Question 29

What is a confusion matrix used for when evaluating a logistic regression classifier?

Accepted Answer

To count TP, TN, FP, and FN predictions

Answer

To scale the features

Answer

To choose the learning rate

Answer

To compute the sigmoid

Question 30

What does it mean if a logistic regression model has very high training accuracy but low test accuracy?

Accepted Answer

The model is overfitting

Answer

The model is underfitting

Answer

The data is perfectly clean

Answer

The threshold is wrong

Test yourself