Model Shot Success by Location
Company: Google
Role: Data Scientist
Category: Machine Learning
Difficulty: hard
Interview Round: Technical Screen
You need to build a model that predicts the probability that a shot becomes a goal for every location on a soccer field. Assume you have historical shot-level data with coordinates on the pitch and other pre-shot context. Describe how you would:
- define the prediction target,
- choose useful features,
- select an appropriate model,
- handle sparse regions of the field,
- validate performance and calibration,
- avoid leakage,
- and present the final output as a spatial heatmap with uncertainty.
Be explicit about trade-offs between interpretability, predictive power, and calibration.
Quick Answer: This question evaluates a candidate's skill in spatial probabilistic modeling, feature engineering, calibration, uncertainty quantification, and handling data leakage when predicting shot success across a soccer pitch.