Exploratory data analysis introduction

Exploratory data analysis (EDA) is detective work on a dataset before modeling: understand shape, spot errors, form hypotheses, and decide what to clean. EDA is iterative—each plot or summary may send you back to the data dictionary.

Goals of EDA

• Understand structure — rows, columns, types, keys
• Assess quality — missing values, duplicates, impossible ranges
• Summarize distributions — center, spread, skew
• Explore relationships — correlations, segments, time trends
• Generate hypotheses — what might predict the outcome?

EDA does not prove causation—it prepares trustworthy questions for modeling and stakeholders.

Typical EDA order

1. Read the data dictionary and business context
2. Count rows/columns; list column types
3. Profile missingness and duplicates
4. Summarize numeric columns (mean, median, quantiles)
5. Tabulate categorical columns (counts, proportions)
6. Plot or cross-tab relationships worth investigating

With Pandas locally, df.info(), df.describe(), and df.groupby() accelerate these steps; this track starts with Python lists and dicts in the playground.

Questions to write down

Before opening tools, answer on paper:

• What is one row? (user, order, session?)
• What is the target or KPI?
• What time range and filters apply?
• What would surprise a domain expert?

Connect to SQL

Many teams EDA in two layers: aggregate in SQL (counts, daily rollups), then load a sample into Python for deeper stats. The warehouse answers “how big”; the notebook answers “what pattern.”

Important interview questions and answers

1. Q: EDA vs modeling?
   A: EDA explores and questions data; modeling fits patterns to predict or explain with stated assumptions.
2. Q: Why EDA before cleaning?
   A: You cannot impute or drop wisely until you know how missingness and outliers are distributed.

Self-check

1. Name three goals of EDA.
2. Why document the unit of observation before plotting?
3. Where might SQL fit in an EDA workflow?

Tip: EDA is iterative—expect to revisit cleaning after plots.