Posterior probability given model accuracy
Company: Meta
Role: Data Scientist
Category: Analytics & Experimentation
Difficulty: easy
Interview Round: Onsite
Scenario: Security classification model has symmetric 95 % accuracy, base‑rate bad users 5 %. Compute posterior probability a user is bad when flagged.
Question 1: If 5 % of users are bad and model accuracy is 95 % on both classes, what is P(true bad | predicted bad)? (Hint: Bayes’ theorem)
Quick Answer: This question evaluates a candidate's understanding of Bayesian inference and conditional probability as applied to binary classifiers, with emphasis on interpreting sensitivity, specificity and base-rate effects.