Method¶

https://www.reddit.com/r/MachineLearning/comments/ob2rll/d_how_often_is_lightgbm_used_for_forecasting_in/

statistics¶

• Prophet and ARIMA-type models are great for forecasting trends with steady growth and stable seasonality

• but struggle with time-series that have complex dependencies on exogenous variables (e.g. ‘shocks’ relating to weather)

ensemble¶

Use an ensemble of

• Prophet to capture the underlying seasonality, and

• XGBoost to capture the weather/seasonality shock impact