DS Lifecycle Project Flow Q&A
Beginner

Data Science Lifecycle – Interview Q&A

Explain how a DS project runs from idea to production, using frameworks like CRISP‑DM or OSEMN.

1 What are the typical stages in the Data Science lifecycle? easy
Answer: A common view is: 1) business understanding, 2) data collection and understanding, 3) data cleaning and preparation, 4) modeling, 5) evaluation with stakeholders, and 6) deployment and monitoring. Frameworks like CRISP‑DM and OSEMN describe the same idea with slightly different names.
CRISP‑DM OSEMN
2 Briefly describe CRISP‑DM. medium
Answer: CRISP‑DM stands for Cross‑Industry Standard Process for Data Mining with six phases: Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation, and Deployment. It emphasises that the process is iterative—you often go back to earlier phases as you learn.
3 At which stages is stakeholder communication most important? medium
Answer: Communication is critical at the start (to define the right business problem), before deployment (to validate that metrics make sense), and after deployment (to explain results and decide next steps). Good Data Scientists keep stakeholders in the loop throughout, not just after the model is trained.
4 Why does lifecycle usually start with business understanding?easy
Answer: Because model quality is useless if it solves the wrong problem. Early alignment defines success metrics, constraints, and expected business action.
5 What happens in the data understanding phase?easy
Answer: Teams profile sources, check schema/coverage, detect leakage risk, measure missingness, and validate whether available data can actually answer the business question.
6 What is the output of data preparation?medium
Answer: A reproducible training-ready dataset (with transformations, feature definitions, and split strategy documented) plus quality checks.
7 Why create a baseline model before complex models?medium
Answer: Baselines set a minimum acceptable benchmark and reveal whether advanced modeling is truly adding value over simple heuristics.
8 What is the difference between validation and final evaluation?medium
Answer: Validation guides model selection during iteration; final evaluation is a held-out unbiased check before deployment decisions.
9 What does deployment include beyond “put model in API”?medium
Answer: It includes integration with business workflows, rollback strategy, logging, monitoring, retraining plan, versioning, and access/security controls.
10 Why is post-deployment monitoring mandatory?easy
Answer: Data and behavior drift over time. Monitoring catches quality drops, bias shifts, and system failures before they cause major business impact.
11 What is concept drift?medium
Answer: Concept drift happens when the relationship between inputs and target changes, so the model’s learned mapping becomes outdated.
12 Where do experiments/A-B tests fit in lifecycle?medium
Answer: Usually in evaluation and post-deployment phases to verify real-world lift and safely compare model-driven decisions vs control behavior.
13 What documentation should lifecycle produce?easy
Answer: Problem statement, data dictionary, feature definitions, modeling assumptions, evaluation results, deployment runbook, and monitoring/retraining criteria.
14 Why is reproducibility a lifecycle concern?medium
Answer: Without reproducibility, teams cannot reliably debug, audit, retrain, or explain model behavior. Reproducibility protects quality and governance.
15 One-line lifecycle summary for interviews?easy
Answer: Start with business goals, turn data into reliable features, build/evaluate models against both ML and business metrics, deploy safely, and continuously monitor and improve.
Previous Next