Model evaluation: cross-validation, metrics
Machine Learning · Engineering
Study notes
Cancer screening: model predicts 90/100 correctly (accuracy 90%), but misses 8 of 10 cancers (recall 20%): dangerous. Precision 95% means few false alarms. F1 balances. 5-fold CV trains on 4/5, tests on 1/5, rotating: robust estimate. Choose metrics matching costs: recall for disease, precision for spam.