如题

问题如下：

After running his model on the test set, Quinn produces a confusion matrix for evaluating the performance of the model (Exhibit 3). He reminds Wu that since the number of defaults in the dataset is likely much smaller than the number of non-defaults, this needs to be considered in evaluating model performance.

Using Exhibit 3 and Quinn’s reminder, the most appropriate measure of the accuracy of the model is:

选项：

A.

0.79

B.

0.86

C.

0.92

解释：

B is correct. Quinn reminds Wu that there are likely unequal class distributions in the dataset, making F1, the harmonic mean of precision and recall, a better measure of accuracy.

Ÿ Precision, P, measures what proportion of positive identifications were actually correct, where

ü P = (TP)/(TP + FP), TP = True positives, and FP = False positives.

ü P = (TP)/(TP + FP) = 118/(118 + 32) = 0.7866 = 0.79.

Ÿ Recall, R, measures what proportion of actual positives were identified correctly, where

ü R = (TP)/(TP + FN) and FN = False negatives.

ü R = (TP)/(TP + FN) = 118/(118 + 8) = 0.9365 = 0.94.

Ÿ F1 is the harmonic mean of precision and recall and is equal to (2×P×R)/(P + R).

ü F1 = (2×P×R)/(P + R) = (2×0.79×0.94)/(0.79 + 0.94) = 0.86.

A is incorrect. Calculating precision results in 0.79: P = (TP)/(TP + FP) = 118/(118 + 32) = 0.79.

C is incorrect. Accuracy is the percentage of correctly predicted classes out of all predictions:

Ÿ A = (TP + TN)/(TP + FP + TN + FN) = (118 + 320)/(118 + 32 + 320 + 8) = 0.92.

Ÿ When the class distributions in the dataset are unequal, as Wu indicates, F1 is a better measure of the accuracy of the model.