Model Soccer Shot Conversion
Company: Google
Role: Data Scientist
Category: Machine Learning
Difficulty: hard
Interview Round: Technical Screen
You are given event-level soccer shot data, and possibly tracking or contextual data. Build a model that predicts the probability that a shot becomes a goal for any location on the field.
In your answer, specify:
- the prediction target and unit of analysis
- useful features from location, geometry, game state, and player context
- model choices and why
- how to generate a shot-probability map over the field
- offline evaluation metrics, calibration strategy, and data-splitting method
- key sources of bias, leakage, and sparse-data issues
Quick Answer: This question evaluates probabilistic predictive modeling, spatial-temporal feature engineering, model calibration and evaluation, and identification of bias, leakage, and sparse-data issues in event-level sports data within the Machine Learning domain.