Requirement 1. Complete the confusion matrices for the validation set.

Start by constructing a confusion matrix for the cutoff point

0.70.

		Confusion Matrix (0.70)
		Predicted Outcomes
		Buyers	Non-Buyers	Total
Actual	Buyers	48		160
Outcomes	Non-Buyers			640
	Total	168		800

Part 2

Construct a confusion matrix for the cutoff point

0.30.

		Confusion Matrix (0.30)
		Predicted Outcomes
		Buyers	Non-Buyers	Total
Actual	Buyers	128		160
Outcomes	Non-Buyers			640
	Total	640		800

Part 3

Requirement 2. A team of management accountants at

Valley

TV estimates the payoffs from their actions. For every customer it targets,

Valley

TV will spend

$35

to market to that customer. For every smart TV it sells,

Valley

TV makes a profit of

$180

after taking into account the

$35

it spends on that customer. Construct the payoff matrix and determine which cut off value

Valley

TV should use.

Start by constructing the payroll matrix. (Complete all answer boxes. Enter a 0 for any zero balances. Enter a spend or loss amount with a minus sign or parentheses.)

Payoff Matrix
		Predicted Outcomes
		Buyers	Non-Buyers
Actual	Buyers
Outcomes	Non-Buyers

Part 4

Calculate the payoff at each cutoff probability.

Total payoff at cutoff 0.70 =
Total payoff at cutoff 0.30 =

Part 5

Which cut off value should

Valley

TV use?

Valley

TV should use the cutoff value at

0.30

0.70

because it results in the

larger

smaller

predicted payoff.

Part 6

Requirement 3. Are there any other factors Valley

TV should consider before building such a model?

A.

Valley

TV should consider whether or not they could make their ROC curve less predictive.

B.

Valley

TV should consider the potential side effects of marketing too aggressively. If the company's marketing efforts irritate or offend interested customers, they may turn those customers away and lose sales.

C.

Valley

TV should consider whether they could increase the Gini impurity of the data set used to make the prediction models.

D.

No other factors need to be considered beyond those already analyzed when producing the prediction models discussed above.