Predict driver acceptance
Company: Uber
Role: Data Scientist
Category: Machine Learning
Difficulty: medium
Interview Round: Onsite
Design an end-to-end machine learning approach to predict **driver acceptance probability** in a ride-sharing or delivery marketplace.
Assume that when a job offer is shown to a driver, the platform wants to estimate the probability that the driver will accept it. Discuss:
- how you would define the prediction target and observation unit,
- what training data and features you would use,
- how you would handle leakage, delayed labels, and sample-selection issues,
- which baseline and production models you would consider,
- how you would evaluate model quality offline and online,
- how calibration, drift, fairness, and marketplace feedback loops affect deployment,
- and how the prediction would be used in ranking, dispatch, or incentive decisions.
Your answer should cover both modeling and product decision-making, not just algorithm choice.
Quick Answer: This question evaluates a candidate's competency in designing and operationalizing an end-to-end machine learning solution for predicting driver acceptance, covering target and observation-unit definition, feature and training-data design, leakage and delayed-label handling, model choice and calibration, evaluation, fairness, and integration with marketplace decisioning. Commonly asked in Machine Learning and Data Science interviews for production-focused roles, it assesses real-world system design and product-thinking skills and probes both conceptual understanding and practical application across modeling, data engineering, evaluation, and deployment in marketplace domains.