Why are the diabetes features scaled around zero?

The dataset's features were mean-centered and scaled by the dataset's authors, so values such as BMI are not in their original units. That is fine for learning, but it is a reminder to always check how a dataset was prepared.

Data Science with Python Module 4: Exploratory Data Analysis and Visualization

Introduction

Exploratory data analysis, or EDA, is how you get to know a dataset before modeling it. You check the shape of each variable, look for outliers, and see which variables move together. Charts do much of the work. This module uses a dataset bundled with scikit-learn, so no download is needed.

Prerequisites

Module 3 completed

Step-by-Step Guide

Step 1: Load a built-in dataset

scikit-learn includes small real datasets. The diabetes dataset has ten numeric features for 442 patients and a numeric target measuring disease progression. as_frame returns pandas objects.

from sklearn.datasets import load_diabetes

data = load_diabetes(as_frame=True)
df = data.frame
print(df.shape)
df.head()

Step 2: Summarize each variable

describe gives count, mean, spread, and quartiles. Compare the mean with the median (50 percent) to spot skew, and look at the minimum and maximum for outliers. For text columns, value_counts shows how often each category appears.

df.describe().T
df["target"].describe()
# for a categorical column:
# df["city"].value_counts(normalize=True)

Step 3: Plot distributions

A histogram shows how values are spread, and a box plot shows the median, quartiles, and outliers. Always label axes and give the chart a title so it can be understood on its own.

import matplotlib.pyplot as plt

fig, axes = plt.subplots(1, 2, figsize=(10, 4))
df["target"].plot.hist(bins=25, ax=axes[0], title="Disease progression")
df[["bmi", "bp", "s1"]].plot.box(ax=axes[1], title="Selected features")
plt.tight_layout()
plt.show()

Step 4: Look at relationships

A scatter plot shows how two variables relate, and a correlation matrix summarizes all pairs at once. Correlation measures straight-line association from minus one to one. It never proves one thing causes another.

corr = df.corr(numeric_only=True)["target"].sort_values(ascending=False)
print(corr)

df.plot.scatter(x="bmi", y="target", alpha=0.5, title="BMI vs progression")
plt.show()

Step 5: Compare groups and write down findings

Group a variable into bands and compare the target across them. Finish EDA by listing what you found in plain sentences: which variables matter, what looks wrong, and what to try next. Those notes guide modeling and become part of your final report.

df["bmi_band"] = pd.qcut(df["bmi"], q=4, labels=["low", "mid-low", "mid-high", "high"])
print(df.groupby("bmi_band", observed=True)["target"].agg(["mean", "median", "count"]))

FAQs

Why are the diabetes features scaled around zero?: The dataset's features were mean-centered and scaled by the dataset's authors, so values such as BMI are not in their original units. That is fine for learning, but it is a reminder to always check how a dataset was prepared.

Summary

EDA means summarizing every variable, plotting distributions, checking relationships with scatter plots and correlations, comparing groups, and writing your findings down before you model.