InterviewSkill
Analytical thinking, metrics, cleaning, and interpretation questions for data roles.
EDA is the process of understanding data before modeling or reporting.
It includes checking schema, distributions, missing values, outliers, relationships, and surprising patterns to guide next steps.
Understand why it is missing, then choose deletion, imputation, or modeling strategies.
Missingness can be random or systematic. The method should preserve signal and avoid biasing downstream analysis.
A good metric is aligned with the business goal, measurable, reliable, and hard to game.
Metrics should reflect user or business value and include guardrails so optimization does not damage another important outcome.
It compares groups of users or records that share a starting condition.
Cohorts help reveal behavior over time, such as retention by signup month, without mixing users from different lifecycle stages.
Check whether it is an error, rare valid event, or meaningful signal.
Outliers can come from data quality issues, system changes, fraud, seasonality, or real extremes. Treatment depends on the goal.