Valeriy’s Substack

Valeriy’s Substack

Still Using predict_proba? Here’s Why Your Probabilities Are Lying to You

Scikit-learn’s most popular method doesn’t return probabilities. It returns scores that look like probabilities. The difference will cost you.

Valeriy Manokhin's avatar
Valeriy Manokhin
Mar 23, 2026
∙ Paid

Scikit-learn’s predict_proba doesn’t return probabilities.

It returns scores. Scores that sum to one, live between 0 and 1, and look exactly like probabilities. They’re not.

This isn’t a minor technical footnote. If you’re using predict_proba outputs to make decisions — setting thresholds, ranking customers, pricing risk, triaging patients — you are build…

User's avatar

Continue reading this post for free, courtesy of Valeriy Manokhin.

Or purchase a paid subscription.
© 2026 Valery Manokhin · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture