Question

Evaluating Predictive Performance ( Business Analytics)

Confusion Matrix !!!

2. A publisher plans to boost the sales of its most popular magazine by sending out promotional mails. We refer to a customer as a responder if he/she subscribes to the magazine for the next year after receiving a promotional mail. Otherwise the customer is referred to as a non-responder. Denote responder by C1 and non-responder by C2. The publisher has built a model to classify each customer as either a responder or a non-responder. In practice only 1% of the customers are responders, and the remaining 99% are non-responders. In order to build an unbiased model, the publisher employed the oversampling method in creating the training set and the validation set, such that both datasets contain 50% responders and 50% non-responders. Please adjust the following validation confusion matrix for oversampling, then compute accuracy measures using the adjusted confusion matrix. Keep at least three digits after the decimal point.

		Actual Class
		C1	C2
Predicted Class	C1	645	112
	C2	255	788

Oversampling factor of C1:
Oversampling factor of C2:

Adjusted validation confusion matrix:

		Actual Class
		C1	C2
Predicted Class	C1
	C2

Overall Accuracy
Overall Error Rate
False Discovery Rate (FDR)
False Omission Rate (FOR)
Precision
Specificity
Sensitivity

Answer 1

PAGE No. DATE Actual Class Given thot C i Tp 64S 452. Prredicted c class goo 645 + 88 64S" +チ8 8 + 112 t-2ss-- 019 61 1483 800 FP + FN 36子 180 0 o.2 O38 ③False Discovery Rate (FDR) FP 11 2 112 + 645-75구 112 0.1419 G False omission Rate CFoR) FN 2SJ -O 244

PAGE No. DATE S Psecision Tp Tp+Fp 64S 64S" + 112" 6 衫子 6 7-88 788 TP TPHEN 64S 64S"十255-960 (7 64S 43166