Exploratory Data Analysis (EDA), Step by Step

What EDA is, and why it has a name

You already know the individual moves — head, info, describe, groupby, charts. Exploratory Data Analysis (EDA) is doing them on purpose, in order, the very first thing on any new dataset. The goal is not a finished answer; it is to understand the data and form questions before you clean hard or build a model. Skipping EDA is how people build charts on broken data and trust models that were doomed from the start.

EDA follows a simple loop. We will walk each step on a small employee table.

1. Look at the whole — size, columns, types, missing values (shape, info).
2. Summarise the numbers — typical values and spread (describe).
3. Univariate — study one column at a time (its distribution).
4. Bivariate — study two columns together (do they relate?).

import pandas as pd

df = pd.DataFrame({
    'dept':       ['Eng', 'Eng', 'Sales', 'Sales', 'Eng', 'Sales'],
    'experience': [5, 2, 8, 3, 10, 6],
    'salary':     [90, 60, 85, 45, 130, 70]
})

Step 1–2: the big picture and the number summary

describe() gives a one-shot summary of every numeric column: how many values, the average, the spread, and the smallest/largest. It is the fastest way to feel a dataset.

print(df.describe())

Step 3: univariate analysis (one column at a time)

Univariate means “one variable”. You look at a single column on its own to learn its distribution — where values cluster, and whether any outliers lurk. A histogram is the classic tool.

import matplotlib.pyplot as plt

df['salary'].plot(kind='hist', bins=5, title='Salary distribution')
plt.xlabel('Salary (thousands)')
plt.show()

Step 4: bivariate analysis (two columns together)

Bivariate means “two variables”. Now you ask whether two columns relate. The quickest numeric tool is the correlation matrix; the quickest picture is a heatmap, where colour shows the strength of each correlation.

import seaborn as sns
import matplotlib.pyplot as plt

corr = df[['experience', 'salary']].corr()
sns.heatmap(corr, annot=True, cmap='coolwarm')
plt.title('Correlation heatmap')
plt.show()