K-nearest neighbors quiz

Question 1

What does the 'K' in K-nearest neighbors represent?

Accepted Answer

The number of nearest neighbors considered for a prediction

Answer

The number of features in the dataset

Answer

The number of output classes

Answer

The learning rate of the model

Question 2

Why is KNN called a 'lazy learner'?

Accepted Answer

It defers all computation until prediction time and just stores the data

Answer

It trains very slowly over many epochs

Answer

It only learns from a small subset of data

Answer

It uses gradient descent inefficiently

Question 3

In KNN classification, how is the class of a new point typically decided?

Accepted Answer

By majority vote among the K nearest neighbors

Answer

By averaging all feature values

Answer

By picking the farthest neighbor's class

Answer

By using a probability threshold of 0.5 only

Question 4

What is the Euclidean distance between two points (1, 2) and (4, 6)?

Accepted Answer

5

Answer

7

Answer

25

Answer

3

Question 5

Why is feature scaling important before applying KNN?

Accepted Answer

So features with large ranges do not dominate the distance calculation

Answer

Because KNN cannot handle decimal numbers

Answer

To reduce the number of neighbors needed

Answer

To make training faster

Question 6

Can KNN be used for regression tasks? If so, how does it predict?

Accepted Answer

Yes, by averaging the target values of the K nearest neighbors

Answer

No, KNN only works for classification

Answer

Yes, by taking a majority vote of values

Answer

No, regression requires gradient descent

Question 7

What happens if you choose K = 1 in KNN?

Accepted Answer

The model becomes sensitive to noise and tends to overfit

Answer

The model always underfits

Answer

The model ignores the nearest point

Answer

The model becomes a linear classifier

Question 8

What happens if K is set too large (e.g., equal to the number of training samples)?

Accepted Answer

The model underfits and tends to predict the majority class

Answer

The model overfits to noise

Answer

The model becomes faster and more accurate always

Answer

Distance no longer matters and it picks randomly

Question 9

Why is it often recommended to use an odd value of K in binary classification?

Accepted Answer

To avoid ties in the majority vote between two classes

Answer

To make distance calculations faster

Answer

Because even numbers are not allowed

Answer

To reduce memory usage

Question 10

What is the Manhattan distance and how does it differ from Euclidean distance?

Accepted Answer

Sum of absolute coordinate differences, vs straight-line for Euclidean

Answer

It is the same as Euclidean distance

Answer

It only works in 2D

Answer

It measures angle between vectors

Question 11

What is the 'curse of dimensionality' and how does it affect KNN?

Accepted Answer

In high dimensions distances become similar, making neighbors less meaningful

Answer

It means too few features cause overfitting

Answer

It refers to having too many training samples

Answer

It is the cost of choosing an even K

Question 12

What is the approximate time complexity of making a single prediction with brute-force KNN?

Accepted Answer

O(n * d) per query for distance computation

Answer

O(1) constant time

Answer

O(log n)

Answer

O(d^2) independent of n

Question 13

How can data structures like KD-trees speed up KNN?

Accepted Answer

They prune far-away regions to avoid checking every point

Answer

They compress the data losslessly

Answer

They train a neural network for search

Answer

They convert KNN into a linear model

Question 14

What is distance-weighted KNN and why is it useful?

Accepted Answer

Closer neighbors get more influence, often weighted by 1/distance

Answer

All neighbors get equal weight always

Answer

Only the farthest neighbor votes

Answer

Weights are learned via backpropagation

Question 15

Is KNN a parametric or non-parametric algorithm, and what does that mean?

Accepted Answer

Non-parametric: it makes no fixed assumption about the data's form

Answer

Parametric: it fits a fixed set of weights

Answer

Parametric: it assumes a normal distribution

Answer

Non-parametric: it uses no training data

Question 16

How does the choice of K affect the bias-variance tradeoff in KNN?

Accepted Answer

Small K = high variance, large K = high bias

Answer

Small K = high bias, large K = high variance

Answer

K has no effect on bias or variance

Answer

Both bias and variance always decrease with K

Question 17

What is a common method to choose the best value of K?

Accepted Answer

Cross-validation across several candidate K values

Answer

Always using K equal to the number of classes

Answer

Picking K randomly

Answer

Setting K to the number of features

Question 18

Why does KNN require a lot of memory?

Accepted Answer

It stores the entire training dataset for use at prediction time

Answer

It creates many copies of each feature

Answer

It uses a large neural network

Answer

It stores all possible distance values in advance

Question 19

How does KNN handle a dataset with imbalanced classes, and what problem can arise?

Accepted Answer

Majority-class points dominate the vote, biasing predictions

Answer

It automatically balances the classes

Answer

Imbalance has no effect on KNN

Answer

It always predicts the minority class

Question 20

In scikit-learn, which class is used for KNN classification?

Accepted Answer

KNeighborsClassifier

Answer

KMeansClassifier

Answer

NearestCentroid

Answer

KNNModel

Question 21

What is the difference between KNN and K-means clustering?

Accepted Answer

KNN is supervised classification/regression; K-means is unsupervised clustering

Answer

They are the same algorithm

Answer

KNN is unsupervised; K-means is supervised

Answer

Both are clustering methods

Question 22

Why can KNN naturally model non-linear decision boundaries?

Accepted Answer

It decides locally, so the boundary adapts to data shape

Answer

It fits a single straight line

Answer

It uses a polynomial of fixed degree

Answer

It assumes classes are linearly separable

Question 23

How should categorical features be handled before using KNN with Euclidean distance?

Accepted Answer

Encode them (e.g., one-hot) so distances are meaningful

Answer

Leave them as text strings

Answer

Delete all categorical features

Answer

Multiply them by the number of classes

Question 24

What does it mean that KNN has no explicit training phase?

Accepted Answer

Fitting just stores data; real work happens at prediction time

Answer

It trains for thousands of iterations

Answer

It computes weights using calculus

Answer

It never uses the training data

Question 25

Given neighbors with classes [A, A, B, A, B] for K=5, what does majority voting predict?

Accepted Answer

A

Answer

B

Answer

Tie, no decision

Answer

Both A and B

Question 26

How does the Minkowski distance generalize Euclidean and Manhattan distances?

Accepted Answer

p=1 gives Manhattan and p=2 gives Euclidean distance

Answer

p=0 gives Euclidean distance

Answer

It only equals Manhattan distance

Answer

It is unrelated to either metric

Question 27

Why is KNN sensitive to noisy data and outliers, especially with small K?

Accepted Answer

Few neighbors mean one bad point can flip the prediction

Answer

Outliers are automatically removed

Answer

Noise only affects large K

Answer

KNN ignores all noisy points

Question 28

What preprocessing step using min-max normalization rescales features to a fixed range?

Accepted Answer

(x - min) / (max - min) maps values to [0, 1]

Answer

(x - mean) / max

Answer

x / sum(x)

Answer

log(x) transformation

Question 29

When applying scaling in a KNN pipeline, why must you fit the scaler only on training data?

Accepted Answer

To prevent data leakage from the test set into the model

Answer

Because test data has no features

Answer

To make the scaler run faster

Answer

Because training data is always larger

Question 30

What is one major advantage of KNN compared to many other algorithms?

Accepted Answer

It is simple, assumption-free, and needs no training phase

Answer

It always runs faster than all other models

Answer

It requires very little memory

Answer

It never needs feature scaling

Test yourself