How would you forecast bike demand?
Company: Two Sigma
Role: Data Scientist
Category: Machine Learning
Difficulty: hard
Interview Round: Technical Screen
You are given historical data from a shared city-bike system and asked to predict usage for a specific docking station during the next hour.
Assume you have access to:
- Hourly trip logs
- Current dock inventory and dock capacity snapshots
- Timestamps in the station's local timezone
- Weather data
- Holiday and local event indicators
- Station metadata such as neighborhood and nearby transit stops
Formulate the problem as predicting the number of bike check-outs from one dock in hour t+1.
Describe:
1. How you would define the target and prediction horizon.
2. What features you would engineer.
3. What baseline and more advanced models you would consider.
4. How you would split the data to avoid time leakage.
5. Which evaluation metrics you would use and why.
6. The most likely causes of overfitting in this problem.
7. How you would prevent, detect, and diagnose overfitting.
8. Important edge cases such as cold-start docks, supply constraints, and missing data.
Quick Answer: This question evaluates a candidate's competency in time-series forecasting, feature engineering, model selection, and evaluation for short-horizon demand prediction in a shared-bike system.