subtopic

Leakage-safe final model selection and one-time test

Refit the chosen pipeline on all training data with the chosen threshold; evaluate ONCE on the test set; write a model card. Concepts: Refit; held-out test; model card Project A: Test: cost, Bad recall/precision, ROC-AUC at the chosen threshold; compare with the always-Good baseline. Project B: Test (2018): MAE, peak-hour MAE; compare with lag-168 and the TSO forecast.