Exam 5: Predictive Analytics I: Trees, K-Nearest Neighbors, Naive Bayes,

An automobile finance company analyzed a sample of recent automobile loans to try to determine key factors in identifying borrowers who would be likely to default on their auto loan. The response variable Default equals 1 if the borrower defaulted during the term of the loan and 0 otherwise. The predictor variable AutoDebt% was the ratio (expressed as a percent) of the required loan payments to the borrower's take-home income at the time of purchase. JobTime was the number of years the borrower had worked at their current job at the time of purchase. CredScore was the borrower's credit score at the time of purchase. Below is part of the classification tree the finance company derived from the data collected in the study. Assume they classify those with a default probability estimate of at least .5 as Defaulters. Based on this classification tree, a member of the study sample who had a credit score of 698, had just started a new job, took out a loan with payments equaling 7% of their income, and defaulted would be

A cable television company has randomly selected a sample of 637 Basic package customers for a marketing test to see which customers are more likely to upgrade to the Premium package. They monitored several predictor variables based on customer activity during the most recently billed month. Then they included a special upgrade offer for the Premium package along with their bill. The response variable Upgrade equals Yes if a customer accepted the offer to upgrade to the Premium package and equals No otherwise. Below is the confusion matrix from the k-nearest neighbors classification analysis. What is the upgrader misclassification rate for the k-nearest neighbors classification?

An automobile finance company analyzed a sample of recent automobile loans to try to determine key factors in identifying borrowers who would be likely to default on their auto loan. The response variable Default equals 1 if the borrower defaulted during the term of the loan and 0 otherwise. The predictor variable AutoDebt% was the ratio (expressed as a percent) of the required loan payments to the borrower's take-home income at the time of purchase. JobTime was the number of years the borrower had worked at their current job at the time of purchase. CredScore was the borrower's credit score at the time of purchase. Below is part of the classification tree the finance company derived from the data collected in the study. Assume they classify those with a default probability estimate of at least .5 as Defaulters. Based on this classification tree, a member of the study sample who had a credit score of 774, just started their current job, took out a loan with payments equaling 19% of their income, and did not default would be

An automobile finance company analyzed a sample of recent automobile loans to try to determine key factors in identifying borrowers who would be likely to default on their auto loan. The response variable Default equals 1 if the borrower defaulted during the term of the loan and 0 otherwise. The predictor variable AutoDebt% was the ratio (expressed as a percent) of the required loan payments to the borrower's take-home income at the time of purchase. JobTime was the number of years the borrower had worked at their current job at the time of purchase. CredScore was the borrower's credit score at the time of purchase. Below is part of the classification tree the finance company derived from the data collected in the study. Assume they classify those with a default probability estimate of at least .5 as Defaulters. A potential borrower with a credit score of 724 who has been at their current job for 6 years is applying for a loan with payments equaling 21% of their income. Based on this classification tree, the best estimate of the probability that this loan applicant would default would be

An automobile finance company analyzed a sample of recent automobile loans to try to determine key factors in identifying borrowers who would be likely to default on their auto loan. The response variable Default equals 1 if the borrower defaulted during the term of the loan and 0 otherwise. The predictor variable AutoDebt% was the ratio (expressed as a percent) of the required loan payments to the borrower's take-home income at the time of purchase. JobTime was the number of years the borrower had worked at their current job at the time of purchase. CredScore was the borrower's credit score at the time of purchase. Below is part of the classification tree the finance company derived from the data collected in the study. Assume they classify those with a default probability estimate of at least .5 as Defaulters. Based on this classification tree, a member of the study sample who had a credit score of 724, been at their current job for 2 years, took out a loan with payments equaling 20% of their income, and did not default would be

The confusion matrix shows the number of observed response variables which are classified correctly.

To predict a qualitative, or categorical, response variable we could use a classification tree.

A regression tree is used for predicting a qualitative response variable.

An MBA admissions officer wishes to predict an MBA applicant's grade point average (GPA) for the MBA program on the basis of the applicant's score on the Graduate Management Admission Test (GMAT) and their undergraduate GPA (UGPA). The admissions officer used a random sample of previously admitted applicants to build a regression tree that can be used to predict the MBA GPAs of future MBA students. Below is the final regression tree. Based on this regression tree, how many of the admitted applicants in the sample had a GMAT score of at least 740?

An MBA admissions officer wishes to predict an MBA applicant's grade point average (GPA) for the MBA program on the basis of the applicant's score on the Graduate Management Admission Test (GMAT) and their undergraduate GPA (UGPA). The admissions officer used a random sample of previously admitted applicants to build a regression tree that can be used to predict the MBA GPAs of future MBA students. Below is the final regression tree. The school awards a Dean's Scholarship to admitted applicants who it predicts will earn a GPA of 3.85 or higher in the MBA program. An MBA applicant has an undergraduate GPA of 3.91. Based on this regression tree, which of the following GMAT scores is the lowest this applicant can earn to qualify for the Dean's Scholarship?

The optimal value of k to use for the k-nearest neighbors approach to predicting a quantitative response variable is the value of k that minimizes RMSE (the square root of the mean of the squared deviations of the predicted values from the observed values).

An internet service provider (ISP) has randomly selected a sample of 223 observations concerning values of the response variable Churn and several predictor variables based on customer activity during the most recently billed month. Here Churn equals Yes if a customer churned-left the internet service provider for another ISP-and equals No otherwise. The predictor variable MinutesOn is the average daily minutes the customer spent online. EmailSent is the average daily number of emails the customer sent from the email address provided by the ISP. ServCalls is the number of times the customer called for service. Below is part of the classification tree they derived from the data collected in the study. Of the sampled customers who spent an average of at least 511 minutes online per day and sent an average of less than 7 emails per day from their ISP-provided email address, what is the sample proportion of those who did not churn?

An automobile finance company analyzed a sample of recent automobile loans to try to determine key factors in identifying borrowers who would be likely to default on their auto loan. The response variable Default equals 1 if the borrower defaulted during the term of the loan and 0 otherwise. The predictor variable AutoDebt% was the ratio (expressed as a percent) of the required loan payments to the borrower's take-home income at the time of purchase. JobTime was the number of years the borrower had worked at their current job at the time of purchase. CredScore was the borrower's credit score at the time of purchase. Below is part of the classification tree the finance company derived from the data collected in the study. Assume they classify those with a default probability estimate of at least .5 as Defaulters. Based on this classification tree, a member of the study sample who had a credit score of 423, been at their current job for 4 years, took out a loan with payments equaling 22% of their income, and did not default would be

Which of the following would you find on a classification tree?

Suppose that a bank wishes to predict whether or not an existing holder of its Silver credit card will upgrade, for an annual fee, to its Platinum credit card. To do this, the bank carries out a pilot study that randomly selects 40 of its existing Silver card holders and offers each Silver card holder an upgrade to its Platinum card. Here, the response variable Upgrade equals 1 if the Silver card holder decided to upgrade and 0 otherwise. Moreover, the predictor variable Purchases is last year's purchases (in thousands of dollars) by the Silver card holder, and the predictor variable PlatProfile equals 1 if the Silver card holder conforms to the bank's Platinum profile and 0 otherwise. Below is the classification tree they derived from the data collected in the study. Based on this classification tree, which of the following Silver card holders would the bank classify as a non-upgrader (assuming they classify those with an upgrade probability estimate of at least .5 as upgraders)?

An MBA admissions officer wishes to predict an MBA applicant's grade point average (GPA) for the MBA program on the basis of the applicant's score on the Graduate Management Admission Test (GMAT) and their undergraduate GPA (UGPA). The admissions officer used a random sample of previously admitted applicants to build a regression tree that can be used to predict the MBA GPAs of future MBA students. Below is the final regression tree. The school requires a predicted MBA GPA of at least 3.50 for an admitted applicant to be considered for a Graduate Assistantship. An MBA applicant has a GMAT score of 650. Based on this regression tree, which of the following undergraduate GPAs is the lowest this applicant can earn to be considered for a Graduate Assistantship?

An automobile finance company analyzed a sample of recent automobile loans to try to determine key factors in identifying borrowers who would be likely to default on their auto loan. The response variable Default equals 1 if the borrower defaulted during the term of the loan and 0 otherwise. The predictor variable AutoDebt% was the ratio (expressed as a percent) of the required loan payments to the borrower's take-home income at the time of purchase. JobTime was the number of years the borrower had worked at their current job at the time of purchase. CredScore was the borrower's credit score at the time of purchase. Below is part of the classification tree the finance company derived from the data collected in the study. Assume they classify those with a default probability estimate of at least .5 as Defaulters. A potential borrower with a credit score of 503 who has been at their current job for 4 years and has a monthly income of $4,700 would like to apply for a loan. To be approved for the loan they would need to be classified as a non-Defaulter. Of the following monthly payments, which is the highest this loan could have to be approved for this potential borrower?

The k-nearest neighbors approach can be used to predict

An internet service provider (ISP) has randomly selected a sample of 223 observations concerning values of the response variable Churn and several predictor variables based on customer activity during the most recently billed month. Here Churn equals Yes if a customer churned-left the internet service provider for another ISP-and equals No otherwise. The predictor variable MinutesOn is the average daily minutes the customer spent online. EmailSent is the average daily number of emails the customer sent from the email address provided by the ISP. ServCalls is the number of times the customer called for service. Below is part of the classification tree they derived from the data collected in the study. Of the sampled customers who spent an average of less than 511 minutes online per day and placed fewer than 3 service calls, what is the sample proportion of those who churned?

One approach to avoid overfitting a classification tree is to use a validation data set to identify valid splits and a training data set to train the classification tree on when to stop making splits.