Conceptual Overview: Explore the link between the table of classification probabilities and the Lift Chart. A bank uses a composite score based on the customer's age, average checking account balance, having a mortgage, job status, marital status, and level of education to predict the probability (between 0 and 1) that the customer will default a loan. The bank must choose a cutoff for loan decisions. Customers with predicted probabilities of default below the cutoff will receive loans and those above the cutoff will not. If the bank grants a loan that is later defaulted, we consider it a Class 1 Error and the bank loses the remaining loan balance. On the other hand, if the bank does not grant a loan to a customer who would have paid the loan back, we consider it a Class 0 Error and the bank loses the interest they would have earned on the loan. Thus, both types of errors cost the bank money. The table below lists the customers in order from highest to lowest predicted probability of default (see the "Prob" column). The "Class" column indicates the actual classification of the customer for each row with "1" indicating default and O1 indicating repaid loan. The "Cumul" column counts the cumulative total number of defaults encountered given that row's predicted probability of default. The graph below is known as a "liffchart" and plots the cumulative number of Class 1 events found (in this case defaults) against the total number of cases as the predicted probability of Class 1 events decreases. Lif-charts assess how much better the classification tool is compared to random guessing (the diagonal line). The fursther the lift-chart line is from the diagonal line, the better the classification. Scroll up and down the table and observe the corresponding point plotted in the lift-chart. Also, drag left or right on the lift-chart to observe the corresponding row of the table. tics Activity 1. If only the customers in the top decile are selected, what percentage would actually default? a. 30% b. 50% c. 60% d. 100% 2. Starting with the fewest customers at the top and increasing the number of cases, the classification performs better than random selection, What is fewest number of cases \{other than zero) where the classification model has the same performance as random selection? a. 15 b. 27 c. 32 d. 50 3. For a particular number of cases taken from the top of the list, the "lift" of the classification model can be measured by the number of actual Class t cases divided by the expected number of ciass 1 cases in a random sample. What is the lift of the classification model for 20 cases? a. 10. b. 1.4 C. 4,4 . 6.0