Build a robust ML pipeline

Q: Build a robust ML pipeline

This is a Machine Learning interview question from Flatiron Health for Data Scientist roles. View the full question and solution on PracHub.

Q: How do I approach Machine Learning interview questions?

Machine Learning questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master machine learning interviews.

Question

You inherit an ML pipeline that predicts next-7-day churn for users, but data quality is inconsistent and feature drift is suspected. A) Propose an end-to-end pipeline design covering: temporal data slicing (label window vs feature window), leakage controls (e.g., using only information available up to prediction time), cross-validation scheme appropriate for time series, and a feature store strategy that guarantees training/serving consistency. B) Define offline metrics (e.g., AUC, PR-AUC, calibration error) and business metrics (e.g., uplift in retention from targeted interventions). Specify how you would threshold scores to optimize a cost-sensitive objective with asymmetric costs. C) Describe concrete data quality and drift monitors: missingness rates, schema checks, training-serving skew, and feature drift using PSI/JS divergence with alert thresholds (e.g., PSI > 0.25 severe). Include how to separate drift in covariates from drift in the target due to product changes. D) Detail an online rollout plan: canary scoring, shadow mode, real-time monitoring, rollback triggers, and retraining cadence. Define explicit retraining triggers (e.g., weekly if PSI moderate for two consecutive weeks or business KPI degrades by X%). Address fairness checks across at least two sensitive cohorts and how you would mitigate disparities.

Build a robust ML pipeline

Comments (0)