A fixed set of hand-coded rules that never change

A way for computers to find patterns in data and improve at a task without being explicitly programmed for every case

Grouping customers into segments with no labels provided

Predicting house prices from a dataset of houses with known sale prices

Reducing the number of features in a dataset

Classifying emails as spam or not spam using labeled examples

Predicting a numeric salary from years of experience

Grouping similar customers together without any predefined labels

The average squared distance of each value from the mean

Dataset A's values are more spread out from the mean than dataset B's

Dataset A has fewer values than dataset B

Dataset A and B are identical in distribution

The student scored higher than exactly half of the other students

The student got exactly 50% of the questions right

The student scored in the bottom 10 percent

It has a long tail extending toward higher values

It is perfectly symmetric around the mean

It has a long tail extending toward lower values

To visually inspect whether a relationship exists between two variables

To compute the exact regression coefficients

A strong linear relationship between the variables

The data contains only categorical values

The two variables are completely unrelated

To make the fitted line pass as closely as possible through the data points

There is a strong linear relationship between the variables

There is little to no linear relationship between the variables

One variable perfectly predicts the other

It guarantees the best performance on new, unseen data

R-squared always decreases as degree increases

It can lead to overfitting, where the curve fits noise instead of the true pattern

That feature has little linear effect on the predicted outcome, holding other features constant

That feature is the most important predictor in the model

The model cannot be trained without removing that feature

So that features with larger numeric ranges do not unfairly dominate distance or gradient calculations

So that categorical features can be removed automatically

To get an honest estimate of how the model performs on data it has never seen

To increase the number of features available

It is too simple to capture the underlying pattern, performing poorly even on training data

It memorizes the training data perfectly but fails on new data

It achieves perfect accuracy on both training and test data

A test on a feature that splits the data into branches

The model correctly predicted the negative class

The model incorrectly predicted the positive class

The model correctly predicted the positive class

It catches almost all spam but also flags many legitimate emails as spam

It balances precision and recall, which is useful when classes are imbalanced

It is always easier to calculate than accuracy

It ignores false positives and false negatives entirely

K-means produces a fixed number of clusters chosen in advance, while hierarchical clustering builds a tree of nested groupings

Hierarchical clustering always requires labeled data while k-means does not

K-means always produces more clusters than hierarchical clustering

It outputs a probability that is mapped to a discrete class label

It can only be applied to time-series data

It always produces a value between negative infinity and positive infinity

An S-shaped curve bounded between 0 and 1

A setting chosen before training, such as tree depth, rather than learned from data

A value predicted by the model for new data

The categories have no inherent order, such as colors or country names

The categories can be ranked meaningfully, such as small, medium, and large

It averages predictions from many models trained on different resampled subsets of the data

It trains a single very deep tree on the full dataset

It removes all categorical features before training

It gives a more reliable estimate of model performance by testing on multiple different splits of the data

It eliminates the need for a test set entirely in production

It always trains the model faster than a single split

A high true positive rate combined with a low false positive rate

A high false positive rate combined with a low true positive rate

The classifier performs no better than random guessing

It sets the number of nearby training points consulted to make a prediction

It sets the maximum depth of a decision tree

Machine Learning Quiz