TimesFM 3.0 Outperforms XGBoost in Building Electricity Forecasting

Benchmarking statistical, ML, and foundation models for day-ahead university building electricity demand. TimesFM 3.0 with covariates reduces mean absolute error by 43% compared to seasonal-naive forecasting, though XGBoost remains competitive in specific daily comparisons.