Outliers and anomaly detection quiz

Question 1

What is an outlier in a dataset?

Accepted Answer

A data point far from most other observations

Answer

The most common value in the data

Answer

The average of all values

Answer

A missing value in the dataset

Question 2

How does a single large outlier affect the mean and the median of a dataset?

Accepted Answer

It strongly affects the mean but barely affects the median

Answer

It strongly affects the median but not the mean

Answer

It affects neither

Answer

It affects both equally

Question 3

In the Z-score method for outlier detection, what does a Z-score measure?

Accepted Answer

How many standard deviations a point is from the mean

Answer

The probability a point is correct

Answer

The distance between two points

Answer

The median of the data

Question 4

Using the common 3-sigma rule, what range of values is considered normal for roughly normally distributed data?

Accepted Answer

mean ± 3 standard deviations

Answer

mean ± 1 standard deviation

Answer

median ± 3 standard deviations

Answer

0 to the maximum value

Question 5

What is the Interquartile Range (IQR) and how is it used to detect outliers?

Accepted Answer

Q3 - Q1, with fences at Q1-1.5*IQR and Q3+1.5*IQR

Answer

Q1 + Q3, with fences at the mean

Answer

Max - Min, with fences at the median

Answer

The standard deviation times 3

Question 6

Which parts of a boxplot reveal potential outliers?

Accepted Answer

Points plotted beyond the whiskers

Answer

The line in the middle of the box

Answer

The width of the box

Answer

The color of the box

Question 7

What is the difference between an outlier and an anomaly?

Accepted Answer

An anomaly is usually a meaningful/interesting outlier we want to detect

Answer

They are completely unrelated concepts

Answer

An outlier is always fraud

Answer

An anomaly is always a data-entry error

Question 8

What are the three main types of anomalies in anomaly detection?

Accepted Answer

Point, contextual, and collective anomalies

Answer

Small, medium, and large anomalies

Answer

Linear, quadratic, and cubic anomalies

Answer

Visual, audio, and text anomalies

Question 9

Give an example of a contextual anomaly.

Accepted Answer

A summer-normal temperature occurring in winter

Answer

The largest number in any dataset

Answer

A duplicated row

Answer

A missing value

Question 10

Why is fraud detection usually an imbalanced classification problem?

Accepted Answer

Fraud cases are extremely rare compared to normal cases

Answer

Fraud cases are the majority

Answer

There are equal numbers of each class

Answer

Fraud is impossible to label

Question 11

Why is accuracy a poor metric for evaluating anomaly detection on imbalanced data?

Accepted Answer

It can be high even when no anomalies are detected

Answer

It is always zero for anomalies

Answer

It is the same as recall

Answer

It cannot be computed for rare classes

Question 12

What is the Isolation Forest algorithm's core idea for detecting anomalies?

Accepted Answer

Anomalies are isolated with fewer random splits (shorter paths)

Answer

Anomalies need the most splits to isolate

Answer

It clusters data by distance

Answer

It fits a normal distribution to the data

Question 13

How does the Local Outlier Factor (LOF) method identify anomalies?

Accepted Answer

By comparing a point's local density to its neighbors' densities

Answer

By measuring distance from the global mean

Answer

By counting missing values

Answer

By fitting a straight line

Question 14

How can the DBSCAN clustering algorithm be used for anomaly detection?

Accepted Answer

Points labeled as noise (no cluster) are treated as anomalies

Answer

The largest cluster is the anomaly

Answer

The cluster centers are anomalies

Answer

Every point is an anomaly

Question 15

What is the modified Z-score, and why use the median absolute deviation (MAD)?

Accepted Answer

It uses the median and MAD, which resist the influence of outliers

Answer

It uses the maximum value instead of the mean

Answer

It ignores the data entirely

Answer

It is identical to the ordinary Z-score

Question 16

What is the difference between supervised and unsupervised anomaly detection?

Accepted Answer

Supervised uses labels; unsupervised relies on rarity/structure without labels

Answer

Both always require labels

Answer

Unsupervised needs more labels than supervised

Answer

They are identical

Question 17

What does 'winsorizing' do to outliers?

Accepted Answer

Caps extreme values at chosen percentiles instead of deleting them

Answer

Deletes all outliers permanently

Answer

Doubles the outlier values

Answer

Converts outliers to missing values

Question 18

Should outliers always be removed from a dataset?

Accepted Answer

No, because some outliers are meaningful and important

Answer

Yes, always delete every outlier

Answer

Yes, outliers are never useful

Answer

No, you must never touch outliers

Question 19

How do outliers affect a linear regression model fit by least squares?

Accepted Answer

They can strongly distort the slope and intercept of the line

Answer

They have no effect on the line

Answer

They always improve the fit

Answer

They only affect the R-squared, never the line

Question 20

What is a common cause of outliers in real-world data?

Accepted Answer

Data-entry or measurement errors, or genuine rare events

Answer

Only random number generation

Answer

Always a programming bug in pandas

Answer

Outliers have no cause

Question 21

What is the percentile (or quantile) capping method for handling outliers?

Accepted Answer

Flagging/capping values beyond chosen percentile thresholds

Answer

Removing the most frequent values

Answer

Replacing all values with the mean

Answer

Only keeping the median

Question 22

Why might the standard Z-score method fail on heavily skewed data?

Accepted Answer

It assumes normality, and mean/std are themselves distorted by skew

Answer

It only works on integers

Answer

It requires labeled data

Answer

It ignores the mean entirely

Question 23

What is the difference between outlier detection and novelty detection?

Accepted Answer

Novelty detection trains on clean data and judges new points; outlier detection finds outliers within the data

Answer

They are exactly the same

Answer

Novelty detection needs no data

Answer

Outlier detection always uses labels

Question 24

How does a One-Class SVM detect anomalies?

Accepted Answer

It learns a boundary around normal data; points outside are anomalies

Answer

It clusters data into many groups

Answer

It fits a regression line

Answer

It counts duplicate rows

Question 25

How can an autoencoder neural network be used for anomaly detection?

Accepted Answer

Anomalies have high reconstruction error since the model learned only normal patterns

Answer

Anomalies have zero reconstruction error

Answer

Autoencoders need labeled anomalies

Answer

It only works on text data

Question 26

What is a scatter plot useful for in detecting outliers between two variables?

Accepted Answer

Spotting points far from the main cloud of two-variable data

Answer

Counting the number of categories

Answer

Showing time on a clock

Answer

Measuring text length

Question 27

What does Mahalanobis distance measure, and why is it useful for multivariate outlier detection?

Accepted Answer

Distance from the center accounting for variable correlation and spread

Answer

Simple straight-line distance ignoring correlation

Answer

The number of features

Answer

The mean of all distances

Question 28

In anomaly detection, what is a false positive?

Accepted Answer

A normal point wrongly flagged as an anomaly

Answer

An anomaly correctly detected

Answer

A missing value

Answer

An anomaly that was missed

Question 29

What is a false negative in anomaly detection, and why can it be costly?

Accepted Answer

A real anomaly missed and labeled normal, which can be very costly

Answer

A normal point flagged as anomaly

Answer

A duplicate record

Answer

A correctly detected anomaly

Question 30

Why should outlier detection thresholds usually be applied after splitting data, using only training-set statistics?

Accepted Answer

To avoid data leakage from the test set into preprocessing

Answer

Because it is faster to compute

Answer

Because the test set has no outliers

Answer

It makes no difference at all

Test yourself