PracHub
QuestionsPremiumLearningGuidesCheatsheetNEWCoaches
|Home/Machine Learning/Snapchat

Optimize Churn Prediction: Feature Engineering and Model Selection

Last updated: Mar 29, 2026

Quick Overview

This question evaluates a data scientist's competency in large-scale feature engineering, model selection and tuning, model explainability, and production debugging for a weekly churn prediction pipeline operating on millions of users.

  • hard
  • Snapchat
  • Machine Learning
  • Data Scientist

Optimize Churn Prediction: Feature Engineering and Model Selection

Company: Snapchat

Role: Data Scientist

Category: Machine Learning

Difficulty: hard

Interview Round: Onsite

##### Scenario You own a churn-prediction pipeline that trains weekly on 10M users. ##### Question Walk me through feature engineering, model selection and hyper-parameter tuning for churn prediction. Why might you favor Gradient Boosted Trees over Logistic Regression here? Describe two techniques for explaining model outputs to non-technical stakeholders. If recall suddenly drops by 15% week-over-week, outline a debugging checklist. ##### Hints Discuss imbalance handling, SHAP, feature drift, and offline/online parity checks.

Quick Answer: This question evaluates a data scientist's competency in large-scale feature engineering, model selection and tuning, model explainability, and production debugging for a weekly churn prediction pipeline operating on millions of users.

Related Interview Questions

  • Explain Overfitting and Transformer Attention - Snapchat (medium)
  • Discuss ML Project Tradeoffs - Snapchat (medium)
  • Model an ads ranking system - Snapchat (medium)
  • Explain BatchNorm, optimizers, and L1/L2 - Snapchat (medium)
  • Explain CLIP, contrastive losses, and retrieval limits - Snapchat (medium)
Snapchat logo
Snapchat
Aug 4, 2025, 10:55 AM
Data Scientist
Onsite
Machine Learning
86
0

Weekly Churn Prediction (10M users): Feature Engineering, Model Choice, Explainability, and Debugging

Scenario

You own a weekly churn-prediction pipeline that trains on 10 million users. The goal is to predict who will churn so the business can target retention interventions.

Tasks

  1. Feature Engineering
    • Define the label, observation/prediction windows, and leakage controls.
    • Propose key feature families and how to handle imbalance.
  2. Model Selection and Hyper-parameter Tuning
    • Describe the model development process, evaluation, and tuning strategy at this scale.
  3. Model Choice Rationale
    • Why might you favor Gradient Boosted Trees (GBTs) over Logistic Regression (LR) here?
  4. Explainability
    • Describe two techniques for explaining model outputs to non-technical stakeholders.
  5. Production Debugging
    • If recall drops by 15% week-over-week, provide a step-by-step debugging checklist.

Hints: Discuss imbalance handling, SHAP, feature drift, and offline/online parity checks.

Solution

Show

Comments (0)

Sign in to leave a comment

Loading comments...

Browse More Questions

More Machine Learning•More Snapchat•More Data Scientist•Snapchat Data Scientist•Snapchat Machine Learning•Data Scientist Machine Learning
PracHub

Master your tech interviews with 7,500+ real questions from top companies.

Product

  • Questions
  • Learning Tracks
  • Interview Guides
  • Resources
  • Premium
  • For Universities
  • Student Access

Browse

  • By Company
  • By Role
  • By Category
  • Topic Hubs
  • SQL Questions
  • Compare Platforms
  • Discord Community

Support

  • support@prachub.com
  • (916) 541-4762

Legal

  • Privacy Policy
  • Terms of Service
  • About Us

© 2026 PracHub. All rights reserved.