Lecture 1 - CRISP-DM Overview

Question 1

What does CRISP-DM stand for?

Answer 1

Cross Industry Standard Process for Data Mining

Question 2

What are the phases of CRISP-DM

Answer 2

1. Project understanding 2. Data Understading 3. Data preparation 4. Modeling 5. Evaluation 6. Deployment

Question 3

What is done during Project Understanding phase?

Answer 3

• Problem formulation (objectives, benefits, assumptions).
• Mapping the problem formulation to a data analysis task
• Understanding the situation

Question 4

What does 80-20 mean for Project Understanding and data understanding?

Answer 4

20% of time spent for project and data understanding, but importance for success is 80%.

Question 5

What are the problems for project understanding?

Answer 5

• Communication problems between domain and data analysis experts
• Project owner may not understand
• Analysts often don’t understand the domain well
• Project owner may not understand the produced models or how to use them
• Formulating goals can be difficult

Question 6

How to define the project objective?

Answer 6

Objective: Increase revenues
Deliverable: Software that automatically selects customers for emailing
Success criteria: revenue increases by 5%

Question 7

What does assessing the situation mean?

Answer 7

Finding out the
- requirements and constraints (laws, ethical issues like gender and race, technical constraints)
- assumptions (representativeness, informativeness, data quality, presence of external factors)

Question 8

How to determine analysis goal?

Answer 8

The primary objective must be transformed into a more technical data mining goal.
- Determine data mining tasks (classification, regression, clustering, associations)
- Predictive accuracy?
- Model flexibility
- Interpretability?
- Runtime
- Interestingess and use of expert knowledge
- Should be understandable for the user

Question 9

What are some ethical issues for data analysis?

Answer 9

• Anonymizing is difficult
• Discriminate (sex, religion, race)
• Yet same information is ok for medical applications

Question 10

What are the four risks in European Artificial Intelligence Act (AI Act)?

Answer 10

Minimal risk: Spam filters etc. face no obligations
Specific transparency risk: System like chatbots must inform users that they are ineracting with a machine
High risk: AI system such as AI-based medical software or AI systems used for recruitment must comply with strict requirements
Unacceptable risk: AI systems that allow social scoring are clear threat to people’s fundaamental rights and banned