End-to-End Project: Classifying Flowers

Time to put everything together. The Iris dataset is the most famous starter dataset in machine learning: 150 flowers, 4 measurements each, and 3 species to predict. It comes built into scikit-learn, so it works even offline.

1. Load and explore#

import pandas as pd
from sklearn.datasets import load_iris

iris = load_iris(as_frame=True)
df = iris.frame
df["species"] = df["target"].map(dict(enumerate(iris.target_names)))
print(df.shape)
print(df["species"].value_counts().tolist())
print(df.groupby("species")[["petal length (cm)", "petal width (cm)"]].mean().round(2))

Output:

(150, 6)
[50, 50, 50]
            petal length (cm)  petal width (cm)
species                                        
setosa                   1.46              0.25
versicolor               4.26              1.33
virginica                5.55              2.03

Exploring first already tells us a lot: petal size is very different between species. Setosa petals are tiny. A good model should find that.

2. Split#

from sklearn.model_selection import train_test_split

X = df[iris.feature_names]
y = df["species"]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=7, stratify=y)
print(len(X_train), len(X_test))

Output:

105 45

3. Train a decision tree#

A decision tree learns a flowchart of yes/no questions, like “Is petal length below 2.45 cm?”. It’s easy to understand, which makes it great for explaining a model to people who don’t code.

from sklearn.tree import DecisionTreeClassifier, export_text

tree = DecisionTreeClassifier(max_depth=3, random_state=0)
tree.fit(X_train, y_train)
print(export_text(tree, feature_names=list(X.columns)))

Output:

|--- petal width (cm) <= 0.70
|   |--- class: setosa
|--- petal width (cm) >  0.70
|   |--- petal length (cm) <= 4.95
|   |   |--- petal width (cm) <= 1.70
|   |   |   |--- class: versicolor
|   |   |--- petal width (cm) >  1.70
|   |   |   |--- class: virginica
|   |--- petal length (cm) >  4.95
|   |   |--- petal length (cm) <= 5.05
|   |   |   |--- class: virginica
|   |   |--- petal length (cm) >  5.05
|   |   |   |--- class: virginica

The model found the same thing we saw while exploring: a narrow petal (0.7 cm or less) means setosa, then petal length and width separate the other two. (The last split leads to virginica on both sides: trees sometimes make splits that don’t change the answer.)

max_depth=3 limits the tree to 3 levels of questions. Deeper trees can memorise the training data (overfitting), so limiting depth is a simple safety net.

4. Evaluate#

from sklearn.metrics import accuracy_score, confusion_matrix

pred = tree.predict(X_test)
print("Accuracy:", round(accuracy_score(y_test, pred), 3))
print(confusion_matrix(y_test, pred, labels=iris.target_names))

Output:

Accuracy: 0.978
[[15  0  0]
 [ 0 14  1]
 [ 0  0 15]]

98% on flowers the tree never saw: only one mistake, a versicolor taken for a virginica. Those two species really do overlap in size.

5. Which features mattered?#

importance = pd.Series(tree.feature_importances_, index=X.columns).sort_values(ascending=False)
print(importance.round(3))

Output:

petal width (cm)     0.556
petal length (cm)    0.444
sepal width (cm)     0.000
sepal length (cm)    0.000
dtype: float64

The tree used only the petal measurements. Feature importance is how you explain to a manager why a model decides what it decides.

6. Use it#

flower = pd.DataFrame([[6.1, 2.8, 4.7, 1.2]], columns=X.columns)
print(tree.predict(flower)[0])

Output:

versicolor

You’ve built a real ML project#

The same six steps (explore, split, train, evaluate, explain, use) apply to every project, from predicting customer churn to detecting fraud. The Data Scientist career track goes deeper: statistics, feature engineering, random forests, gradient boosting, tuning and full portfolio projects.

Challenge

Replace the decision tree with LogisticRegression(max_iter=1000) and with KNeighborsClassifier(). Which model gets the best test accuracy?

Summary#

• Explore → split → train → evaluate → explain → use.
• Decision trees learn readable yes/no rules; limit max_depth to avoid overfitting.
• feature_importances_ shows which inputs drove the predictions.