1

Question 1

Question 1
The ACFE, or association of certified fraud examiners, estimates that a typical organization loses
a) 2% of its revenues to fraud each year.
b) 5% of its revenues to fraud each year.
c) 10% of its revenues to fraud each year.
d) 20% of its revenues to fraud each year.

Answer 1

b) 5% of its revenues to fraud each year.

Question 2

Question 2
Which of the following is not typical for fraud
a) Uncommon.
b) Well-considered.
c) Imperceptibly concealed.
d) Static in time.
e) Often carefully organized.

Answer 2

d) Static in time.

Question 3

Question 3
The main motives to commit fraud are
a) pressure.
b) opportunity.
c) rationalization.
d) all of the above.

Answer 3

d) all of the above.

Question 4

Question 4
Which statement is NOT CORRECT?
a) Fraud detection aims at applying detection models on new, unseen observations and assigning a fraud risk to every observation.
b) Fraud investigation is typically conducted by a computer to investigate suspicious cases.
c) Fraud confirmation determines the true fraud label, and might involve additional field
research.
d) Fraud prevention then focuses on preventing fraud from being committed in the future.

Answer 4

b) Fraud investigation is typically conducted by a computer to investigate suspicious cases.

Question 5

Question 5
Which statement is CORRECT?
a) Counterfeit is an imitation intended to be passed off fraudulently or deceptively as genuine. Think about a fake credit card, for example.
b) Identity theft is crime of obtaining personal or financial information about another person for the purpose of assuming that person’s name or identity in order to make transactions or purchases.
c) Health care fraud involves the filing of dishonest health care claims in order to make profit. Examples are billing by health practitioners for care that they never rendered, filing duplicate health care claims for the same service rendered, or billing a non-covered service as a covered service.
d) Click fraud is an illegal practice that occurs when individuals maliciously click on a website’s
click-through advertisements, either banner ads or paid text links, to increase the payable
number of clicks to the advertiser.
e) All statements are correct.

Answer 5

e) All statements are correct.

Question 6

Question 6
Behavioral credit card fraud refers to:
a) individuals obtaining new credit cards from issuing companies by using false personal information, and then spending as much as possible in a short space of time.
b) individuals fraudulently obtaining the details of legitimate cards and subsequently making sales on a ‘Cardholder Not Present’ basis.

Answer 6

b) individuals fraudulently obtaining the details of legitimate cards and subsequently making sales on a ‘Cardholder Not Present’ basis.

a) is application fraud

Question 7

Question 7
The implementation of EMV (Europay, MasterCard and Visa) chips in October 2015 led to a
a) led to an increase in CNP (Cardholder Not Present) fraud.
b) led to a decrease in CNP (Cardholder Not Present) fraud.

Answer 7

a) led to an increase in CNP (Cardholder Not Present) fraud.

Question 8

Question 8
What are important Key Performance Indicators (KPIs) when building analytical fraud models?
a) statistical accuracy.
b) interpretability.
c) operational efficiency.
d) economical cost.
e) regulatory compliance.
f) all of the above.

Answer 8

f) all of the above.

Question 9

Question 9
Fraud analytics data sets are typically
a) highly balanced in terms of the number of fraudulent/non-fraudulent cases.
b) highly imbalanced in terms of the number of fraudulent/non-fraudulent cases.

Answer 9

b) highly imbalanced in terms of the number of fraudulent/non-fraudulent cases.

Question 10

Question 10
Which statement is NOT CORRECT?
a) The classic approach to fraud detection is an expert-driven approach, which builds on the experience, intuition, and business or domain knowledge of one or more fraud analysts.
b) Supervised learning is a type of machine learning that needs a labeled data set of historically observed fraud behavior to learn patterns from. It can be used to both predict fraud as well as the amount of fraud.
c) Unsupervised learning is a type of machine learning that starts from an unlabeled data set and performs anomaly detection to find outlying behavior which might possibly indicate fraudulent behavior.
d) Social network learning learns fraudulent behavior by inspecting Facebook, Twitter and LinkedIn pages.

Answer 10

d) Social network learning learns fraudulent behavior by inspecting Facebook, Twitter and LinkedIn pages.

Question 11

Question 11
Which statement is CORRECT?
a) Rule bases or engines are typically expensive to build, since they require extensive manual input by the fraud experts, and often turn out to be difficult to maintain and manage.
b) Rules have to be kept up to date since fraudsters continuously change their tactics.
c) The firing of a rule requires human follow-up and investigation.
d) All statements are correct.

Answer 11

d) All statements are correct.

Question 12

Question 12
The fraud analytics process model is
a) strictly sequential where each step is followed by the next one.
b) iterative such previous steps may have to be re-considered during the construction of an analytical fraud model.

Answer 12

b) iterative such previous steps may have to be re-considered during the construction of an analytical fraud model.

Question 13

Question 13
The aim of feature engineering is to transform data set variables into features
a) so as to help the analytical models achieve better performance in terms of predictive performance.
b) so as to help the analytical models achieve better performance in terms of interpretability.
c) so as to help the analytical models achieve better performance in terms of either predictive performance, interpretability or both.

Answer 13

c) so as to help the analytical models achieve better performance in terms of either predictive performance, interpretability or both.

Question 14

Question 14
Time is
a) an important feature in fraud detection.
b) not an important feature in fraud detection.

Answer 14

a) an important feature in fraud detection.

Question 15

Question 15
Which statement about the RFM variables in fraud detection is correct?
a) Recency measures how long an event or transaction took place.
b) Frequency is another important feature and counts the number transactions per unit of time.
c) Monetary measures the intensity of the transaction, typically expressed in dollar terms.
d) All statements are correct.

Answer 15

d) All statements are correct.

Question 16

Question 16
When working with time stamps,
a) it makes sense to use the arithmetic mean.
b) it does not make sense to use the arithmetic mean.

Answer 16

b) it does not make sense to use the arithmetic mean.
-> use von mises distribution, which makes more sense on circular data

Question 17

Question 17
We can model a timestamp as a periodic variable by using
a) Gaussian distribution.
b) exponential distribution.
c) Poisson distribution.
d) von Mises distribution

Answer 17

d) von Mises distribution

Question 18

Question 18
A zero or small recency could indicate
a) normal behavior.
b) anomalous behavior.

Answer 18

b) anomalous behavior.

Question 19

Question 19
The recency can be defined as
a) exp(-γ) with t the time-interval between two consecutive transfers.
b) exp(+γt) with t the time-interval between two consecutive transfers.
c) exp(-γt) with t the time-interval between two consecutive transfers.
d) exp(-t) with t the time-interval between two consecutive transfers

Answer 19

c) exp(-γt) with t the time-interval between two consecutive transfers.

Question 20

Question 20
Which statement is CORRECT?
a) The monetary feature represents the intensity of the transaction. It usually corresponds to the amount expressed in dollar terms.
b) The monetary feature can be operationalized in various ways such as the average, minimum/maximum, most recent value or using absolute or relative trends.
c) Deviations in monetary spending on credit card transfers can raise a serious suspicion of fraud.
d) All statements are correct.

Answer 20

d) All statements are correct.

Question 21

Question 21
In most data sets, fraud occurs in typically less than
a) 0.5% of the cases.
b) 5% of the cases.
c) 10% of the cases.
d) 20% of the cases.

Answer 21

a) 0.5% of the cases.

Question 22

Question 22
Every classifier faced with a skewed data set typically tends to favor the
a) minority class.
b) majority class.

Answer 22

b) majority class.

Question 23

Question 23
When doing undersampling or oversampling, it is important that the test set
a) is treated in a similar way.
b) remains untouched.

Answer 23

b) remains untouched.

Question 24

Question 24
Practical experience shows that the ratio

a) 20% non-fraudsters versus 80% fraudsters is quite commonly used in the industry.
b) 80% non-fraudsters versus 20% fraudsters is quite commonly used in the industry.
c) 50% non-fraudsters versus 50% fraudsters is quite commonly used in the industry.
d) 0% non-fraudsters versus 100% fraudsters is quite commonly used in the industry.

Answer 24

b) 80% non-fraudsters versus 20% fraudsters is quite commonly used in the industry.

Question 25

Question 25
Which statement about SMOTE is NOT CORRECT?
a) It oversamples the minority class by creating synthetic examples.
b) In Step 1 of SMOTE, for each minority class observation, the k nearest neighbors are
determined, in Euclidean sense for example.
c) SMOTE can also be easily combined with oversampling the majority class.
d) Throughout our research, we have found that this SMOTE methods works really good when
dealing with skewed class distributions.

Answer 25

c) SMOTE can also be easily combined with oversampling the majority class.

Question 26

Question 26
The key idea of undersampling, oversampling, SMOTE and its variants is to adjust the class priors to enable the analytical technique to create a meaningful model that discriminates the fraudsters from the non-fraudsters. By doing
a) the class posteriors become biased.
b) the class priors become biased.

Answer 26

a) the class posteriors become biased.