📏 05 — Evaluation and Scaling

Clustering evaluation is harder because there is no target label.

---

Scaling is Essential

Distance-based algorithms are affected by feature scale.

from sklearn.preprocessing import StandardScaler

X_scaled = StandardScaler().fit_transform(X)

---

Silhouette Score

from sklearn.metrics import silhouette_score

score = silhouette_score(X_scaled, labels)
print(score)

Higher is generally better, but interpretation still matters.

---

Cluster Profiling

Always profile clusters:

df.groupby("cluster").agg(
    avg_spend=("spend", "mean"),
    avg_visits=("visits", "mean"),
    count=("cluster", "size")
)

---

Business Validation

Ask:

• Are clusters stable?
• Are they explainable?
• Can the business act on them?
• Are they just artifacts of scale or outliers?

---

Next

➡️ 06-exercises