witfame

Core Concepts & Basics

1. What are the steps in a data analysis project?
Answer: Understand the problem → Gather data → Clean data → EDA → Modeling → Interpret → Share insights.
Example: Analyzing checkout abandonment to identify high drop-off due to hidden delivery fees.

2. How do you handle missing or corrupted data?
Answer: Drop rows, fill with mean/median/mode, regression, or use flag variables.
Scenario: Replaced missing income values using regression based on age and location.

3. What tools and languages do you use?
Answer: SQL, Python, Excel, Tableau, Power BI.
Scenario: Merged customer feedback from Excel, SQL DB, and API using Python, visualized in Tableau.

4. Explain different types of joins.
Answer:

• Inner Join: Only matched records
• Left Join: All left + matched right
• Right Join: All right + matched left
• Full Join: All records with NULLs where no match
  Scenario: Join customers and orders to show purchase patterns.

5. Describe a challenging data project.
Answer: Engagement dropped in mobile app post-update. Found Android 12 crashes spiked.
Impact: Developers fixed the issue, engagement restored.

6. Difference between variance and standard deviation?
Answer: Variance = average of squared deviations; Std Dev = sqrt of variance.
Scenario: Std Dev of $1,200 easier to understand than variance of 1.4M.

7. What is A/B testing?
Answer: Comparing two versions of something (A vs B) to see which performs better.
Scenario: CTA button test increased conversion by 12% with B version.

8. How do you explain findings to non-technical stakeholders?
Answer: Use plain language, visuals, and business outcomes.
Scenario: Used heatmap to show that absence of follow-up emails led to 3x churn.

9. What is data normalization?
Answer: Rescaling data to a common scale (0–1 or standard scores).
Scenario: Normalized income to avoid it dominating clustering results.

10. Difference between structured and unstructured data?
Answer:

• Structured: Tables (SQL, CSV)
• Unstructured: Text, images, video
  Scenario: Combined purchase data (structured) and reviews (unstructured) to improve product insight.

---

Intermediate Topics

11. Supervised vs. Unsupervised Learning
Answer: Supervised uses labeled data (e.g., regression); Unsupervised finds patterns (e.g., clustering).
Scenario: Used K-means for customer segmentation.

12. What is overfitting and how to prevent it?
Answer: Overfitting = too complex model fits training data but not new data. Prevent using cross-validation, regularization.
Scenario: Reduced complexity in sales forecast model.

13. Correlation vs. Causation
Answer: Correlation: variables move together. Causation: one causes the other.
Scenario: Ice cream sales & drowning both rise in summer – not causal.

14. What is a p-value?
Answer: The probability that results occurred by chance.
Scenario: In an A/B test, p = 0.03 → significant result.

15. ETL vs. ELT
Answer:

• ETL: Transform before loading
• ELT: Transform after loading (used in modern data lakes)
  Scenario: Used ELT in BigQuery for efficient cloud processing.

16. What is a data warehouse?
Answer: Centralized storage for integrated data, used for analytics.
Scenario: Used Redshift to report on KPIs from sales, marketing, and product.

17. Dimensionality Reduction
Answer: Reducing number of features (e.g., PCA, t-SNE).
Scenario: PCA reduced 25 variables to 3 principal components.

18. Hypothesis Testing
Answer: Used to test if a claim about a dataset is statistically significant.
Scenario: Tested if campaign increased revenue per user (p < 0.05).

19. Time Series Analysis
Answer: Analyzing data over time intervals.
Scenario: Used ARIMA to forecast next quarter’s sales.

20. Confidence Interval
Answer: Range where the true value lies with certain confidence.
Scenario: 95% CI for average time on site = 4.5 to 6.3 minutes.

---

SQL, Statistics, and Logic

21. What is multicollinearity?
Answer: High correlation between predictors → unreliable regression coefficients.
Scenario: Removed one of two correlated marketing budget variables.

22. Primary Key vs. Foreign Key
Answer:

• Primary = Unique record identifier
• Foreign = References another table
  Example: customer_id in Customers is primary; in Orders, it’s foreign.

23. What is data cleaning?
Answer: Process of correcting/removing corrupt or inaccurate records.
Scenario: Removed duplicates, fixed date formats before analysis.

24. What are outliers? How do you handle them?
Answer: Unusually distant values. You can remove, cap, or analyze separately.
Scenario: Removed 100x outlier due to logging error.

25. UNION vs. UNION ALL
Answer: UNION removes duplicates, UNION ALL keeps them.
Scenario: Merged two months of sales data using UNION ALL.

26. Cohort Analysis
Answer: Grouping users by time-based activity for behavioral patterns.
Scenario: Measured retention rate by user signup month.

27. Rolling Average
Answer: Smooths fluctuations by averaging over a time window.
Scenario: Used 7-day rolling average for DAU.

28. KPI (Key Performance Indicator)
Answer: Metric used to evaluate success.
Example: Conversion rate, CLV, CAC, churn rate.

29. Z-score
Answer: Number of standard deviations a value is from the mean.
Scenario: Used to identify server anomalies.

30. Confusion Matrix
Answer: Table used to describe classification performance: TP, FP, TN, FN.
Scenario: Evaluated precision and recall for fraud detection model.

---

Model Evaluation & Pandas

31. Cross-validation
Answer: Validating model by splitting data into multiple folds.
Scenario: Used 5-fold CV for churn prediction.

32. Type I vs. Type II Errors
Answer:

• Type I: False positive
• Type II: False negative
  Scenario: Type I = legit flagged as fraud, Type II = fraud not flagged.

33. Anomaly Detection
Answer: Identifying data points that deviate significantly.
Scenario: Flagged login spikes as potential attacks.

34. Central Limit Theorem
Answer: Sampling distribution of the mean tends toward normal.
Scenario: Justified bootstrapping for confidence intervals.

35. Data Wrangling
Answer: Cleaning + structuring raw data for analysis.
Scenario: Combined multiple tables into single dataset for analysis.

36. Linear Regression Assumptions
Answer: Linearity, independence, homoscedasticity, normality, no multicollinearity.
Scenario: Verified assumptions before final model.

37. Data Visualization Types
Answer: Line chart, bar chart, box plot, heatmap, histogram, etc.
Scenario: Used line chart for trend, heatmap for correlation.

38. Seasonality Detection
Answer: Use autocorrelation, decomposition, and visuals.
Scenario: Found weekly seasonality in orders.

39. .loc vs .iloc in Pandas
Answer: .loc = label-based, .iloc = index-based.
Example: df.loc['2023-01-01'] vs df.iloc[0].

40. count() vs value_counts() in Pandas
Answer:

• count(): non-null count
• value_counts(): frequency of unique values
  Scenario: Used value_counts() to find top 5 selling products.

---

🔹 Business and Strategy

41. What is a dashboard?
Answer: Visual interface showing key metrics.
Scenario: Built interactive sales dashboard with filters by region.

42. Describe a time you improved a process.
Answer: Automated weekly Excel reports into a Python script, saving 4 hours/week.

43. What are your favorite charts and why?
Answer:

• Line for trends
• Bar for comparison
• Box plot for distribution
  Scenario: Box plot helped identify outliers in sales.

44. How do you prioritize tasks in a data project?
Answer: Based on business impact, urgency, and feasibility.
Scenario: Prioritized user churn over feature usage because of revenue impact.

45. How do you ensure data quality?
Answer: Validations, sanity checks, test cases, data profiling.
Scenario: Wrote automated tests for missing values and invalid IDs.

46. What is a funnel analysis?
Answer: Analyzes steps users take before conversion/drop-off.
Scenario: Used for signup process, found high drop-off on password screen.

47. What is a heatmap?
Answer: Visualizes data values using color.
Scenario: Used to visualize feature correlation in ML.

48. What metrics would you track in an e-commerce dashboard?
Answer: Revenue, conversion rate, cart abandonment, top products, returning users.

49. Describe a time you influenced decision-making.
Answer: Suggested changing email cadence after data showed drop in engagement. Result: +25% open rate.

50. Why should we hire you as a data analyst?
Answer: I combine strong analytical skills, tool expertise, and business sense. Proven experience translating raw data into actionable insights that drive results.