Skip to content
ML Academy

Forecasting

Interview question

  • seniorlevel
  • 12 minsuggested time
  • 2Data scientist, ML engineer

You build one global LightGBM model to forecast daily passengers for 300 routes, 14 days ahead. Features include lag_1, lag_7 and a 7-day rolling mean. Backtest error is excellent. What would you check, and how would you structure the model for a 14-day horizon?

Tests: Lag/rolling feature leakage relative to the horizon, direct vs recursive strategies, exogenous variable availability, and global models across series.

day 24day 25

Sign in to practise: answers are saved before model answers are shown.