subtopic

Cost of errors (cost matrix) and the metric it implies

Write the error-cost table and turn it into the metric you will optimise later. Concepts: False positives vs false negatives; asymmetric costs; cost-based evaluation Project A: Hofmann matrix: approve a Bad applicant = cost 5 (lost principal); reject a Good applicant = cost 1 (lost interest). → optimise Bad recall and total cost, choose the threshold by cost (E13). Project B: Asymmetric too: under-forecasting a peak (blackout risk, expensive spot power) costs more than over-forecasting → look at peak-hour error and under-forecast rate, not only MAE.