FRM Part I · FRM Exam Part I · Machine Learning and Prediction
A fraud model is evaluated on 1,000 transactions. The confusion matrix shows: true positives 30, false positives 20, false negatives 10, true negatives 940. Which statement about accuracy and recall is correct?
Accuracy is 97.0% because 970 of 1,000 transactions are classified correctly, while recall is 75.0% because the model catches 30 of the 40 actual frauds. The 60% alternative is precision, which divides by flagged transactions rather than actual frauds.
- AAccuracy is 97.0% and recall is 75.0%Correct
- BAccuracy is 97.0% and recall is 60.0%
- CAccuracy is 97.0% and recall is 97.9%
- DAccuracy is 75.0% and recall is 97.0%
Explanation
Accuracy = (30+940)/1000 = 97.0%. Recall = TP/(TP+FN) = 30/40 = 75.0%. The 60% option is precision-like (30/50), which uses false positives instead of false negatives. Accuracy looks high because fraud is rare.
Did you get it right without looking?
One question tells you little. A timed set on Machine Learning and Prediction shows your real accuracy, how long you take and where you lose marks.
More Machine Learning and Prediction questions
- A risk analyst fits a single, very deep classification tree to predict loan default. It classifies the training data almost perfectly but pe…
- A bank's credit-default classifier is tested on 200 loans. It flags 50 loans as defaults, of which 40 actually defaulted. In total 60 loans …
- A neural network with many parameters achieves very low error on the training set but substantially higher error on a validation set. Which …
- A model predicts whether a trade is fraudulent. Of 1,000 trades, 20 are actually fraudulent. The model flags 25 trades, of which 15 are actu…
- A lender lowers the probability threshold used to classify a borrower as a predicted defaulter, from 0.50 to 0.30, using the same fitted mod…
- A bank combines 100 decision trees by majority vote to classify counterparties as default or non-default. Compared with a single deep tree, …