Chapter 3: performing the Test Plan and Analyzing the Results

Question 1

What are the 4 main types of data analytics?

Answer 1

Descriptive

Diagnostic

Predictive

Prescriptive

Question 2

What is Descriptive Analytics?

Answer 2

procedures that summarize existing data to determine WHAT HAS HAPPEND IN THE PAST

Question 3

What are common tools in Descriptive Analytics?

Answer 3

Summary statistics (mean, median, etc.)

Data reduction/filtering (fuzzy matching)

Question 4

summary statistics

Answer 4

describes set of data in terms of location, range, shape

helps to quicly see how data is distributed

Question 5

(desriptive ) data reduction or filtering

Answer 5

reduce observations, focus on relevant items –> reducing large set in samller set

helps to isolate high risk items, so can focus on what matters

ex: filtering large vendor list to only include those with transactions over 10k

Question 6

What is Diagnostic Analytics?

Answer 6

exploring current data to determine WHY SOMETHING HAS HAPPENED, typically comparing data to benchmark

Question 7

What are the 4 key methods in Diagnostic Analytics?

Answer 7

Profiling

Clustering

Similarity Matching

Co-occurrence Grouping

Question 8

(diagnostic) profiling

Answer 8

identifies typical behaviour of individual/group by compiling summary statistics
–> COMPARING INDIVIDUALS TO POPULATION

shows typical behaviour of a group –> helps to detect abnormal patterns (for ex: fraud)

Question 9

(diagnostic) clustering

Answer 9

identifies groups that have common underlying characteristics –> reveal hidden relationships

Grouping data into clusters based on natural patterns, without pre-defined labels.
finds groups without prior labels

Question 10

(diagnostic) similarity matching

Answer 10

measures how alike two items are and used to group data in clusters
find things that look alike

for ex: detect suspicious entries or fraud

Question 11

(diagnostic) co-occurence grouping

Answer 11

find items that often appear/happen together
“people who buy x, also buy y”

helps reveal recurring transactional patterns

Question 12

What is Predictive Analytics?

Answer 12

It uses historical data and models to forecast future outcomes.

procedures used to generate a model that can determine WHAT IS LIKELY TO HAPPEN IN FUTURE

Question 13

What are 3 common predictive methods?

Answer 13

Regression

Classification

Link Prediction

Question 14

(predictive methods) regression

Answer 14

A: A statistical model used to predict a number (e.g., sales, income) based on other variables.

predicts a number (ex:sales) show how variables are related

good when, high R2 (close to 1)
statistically significant coeffcients (p<005)

Question 15

(predictive methods) classification

Answer 15

predicts class/category for new observation based on manual identification of classes from previous observations

Question 16

(predictive methods) link prediction

Answer 16

PREDICTS which new connections or relationships are likely to form in network based on existing data

predicts connections between items –> like suggesting friend a social media

Question 17

Q: What is Prescriptive Analytics?

Answer 17

procesdures to identify the best possible options for WHAT SHOULD BE DONE IN THE FUTURE

Question 18

What tools are used in Prescriptive Analytics?

Answer 18

Decision support systems
= helps users make decisions by combining data and analysis to recommend best action

machine learning and AI
= they recommend a course of action + model adapts to new external data

Question 19

Decision support systems

Answer 19

rule based systems, that gather data and recommend actions based on input

for ex: cashflow forecasting and management toolsma

Question 20

(prescriptive) machine learning and AI

Answer 20

learns from data to improve suggestions

helps continuously scanning transactions for example flagging suspicous vendors

Question 21

Q: (Similarity Matching) What is fuzzy matching?

Answer 21

finds text entries that are similar but not exactly the same

technique for detecting suspicious records in imperfect data

Question 22

Q: (Similarity Matching) When do we use fuzzy matching?

Answer 22

uses probability to identify likely simir data
when data has inconsistencies, with imperfect data
like “123 Main St.” vs. “123 Main Street.”

Question 23

Q: (Similarity Matching) What’s the risk of using a high fuzzy matching threshold?

also for low

Answer 23

A: Fewer false positives but more false negatives (you miss real matches).

A: More false positives (many things look matched, but aren’t real).

Question 24

Q: (Classification) What is pruning in decision trees?

Answer 24

remove branches from decision tree to avoid overfitting

model works to good in training data, will not work great in test data

Question 25

Q: (Classification) Why do we prune a decision tree? .

Answer 25

make model simpler, more generalizable to new data

Question 26

Q: (Classification) What is a linear classifier?

Answer 26

A: A model that separates data using a straight line (or flat boundary in higher dimensions) based on input variables.

Question 27

Q: (Classification) Why use linear classifiers?

Answer 27

simple, fast, interpretable results good when data can be separated linearly

Question 28

Q: (Classification) What’s a limitation of linear classifiers?

Answer 28

performs bad when nonlinear or complex relationship between variables

Question 29

Q: (Classification) What is a Support Vector Machine (SVM)?

Answer 29

tries to divido 2 groups, NOT JUST STRAIGHT LINE, is a discriminating classifier that is defined by a seprating hyperplane first finds wides margin and then take the middle of 2 widest margin lines --> middle line

Question 30

Q: (Classification) What does the SVM margin mean?

Answer 30

distance between seprating line (hyperplane) and closest data points from each class

Question 31

Q: (Classification) Why is SVM useful?

Answer 31

A: It works well for both linear and non-linear classification, especially with small datasets and clear class separation.

Question 32

Q: (Classification) What is underfitting?

Answer 32

A: A model that is too simple and misses important patterns in the data (high bias, poor training & testing accuracy).

Question 33

Q: (Classification) What is overfitting?

Answer 33

A: A model that is too complex and fits the training data too well, including the noise (low bias, high variance, poor generalization). fits training data too well will do poor in test data

Question 34

Q: (Classification) How to detect underfitting vs. overfitting?

Answer 34

Underfitting: low training & test accuracy Overfitting: high training but low test accuracy

Question 35

(diagnostic) profiling BENFORDS LAW

Answer 35

early detection of irregularities continuous monitoring enables company to analyze real time data

Question 36

Labeled data

Answer 36

Labeled data = data met bekende uitkomst of antwoord (label) Elke observatie (rij in je dataset) heeft niet alleen invoerwaarden (zoals leeftijd, geslacht, inkomen), maar ook een label, bijvoorbeeld: "ja/nee" "goed/slecht" "1/0" "fraude/niet-fraude" "goedgekeurd/afgewezen" 🔁 → Wordt gebruikt bij supervised learning → Het model leert van deze bekende antwoorden om nieuwe gevallen te voorspellen.

Question 37

unlabeled data

Answer 37

Unlabeled data = data zonder uitkomst/antwoord Je hebt alleen kenmerken of eigenschappen (features), maar geen categorie of waarde die je vooraf kent. 🔁 → Wordt gebruikt bij unsupervised learning → Het model zoekt zelf naar patronen of groepen zonder dat je het goede antwoord geeft.