34. Classification in Python with Logistic Regression

34.3 predict_proba Explained Simply

By Ravindra Bagale · Last updated: 24 October 2026

Why. A hard 0/1 is blunt. Ops teams often want “how sure?” — that is predict_proba.

What. predict_proba(X) returns two columns for binary problems: probability of class 0 and class 1. Default predict picks class 1 when P(class 1) ≥ 0.5. You can choose a different cutoff later if false alarms are costly.

How.

# After clf.fit(...)
proba = clf.predict_proba(X_test)
print("P(class 0), P(class 1):\n", proba.round(3))
print("predict:", clf.predict(X_test))

# Manual threshold example: flag churn if P(1) >= 0.4
p1 = proba[:, 1]
print("custom flags @0.4:", (p1 >= 0.4).astype(int))

What you should see. Probabilities between 0 and 1. Rows near 0.5 are “unsure.” Changing the threshold trades missed churns vs too many alerts — business choice, not magic.