Chapter 3 Notes - Jackie Silverman PDF

Preview

Summary

Jackie Silverman...

Description

●

Target Marketing: directing marketing activities to customers who are most likely to buy ○ Implies selection - some customers identified as more valuable than others and these customers receive special attention ○ Skilled targeting can improve a company's profitability, increasing revenues and decreasing costs ● 2 predictive models in target marketing: ○ Regression analysis: when response variable is continuous, employ regression analysis to make predictions ■ ex: of response variables include sales dollars, sales volume, cost of sales, cost of customer support and CLV ○ Classification: when response variable is categorical ■ Ex: of response variables include whether a customer buys, whether a customer remains with the company or leaves to buy products elsewhere ● Logistic regression model provides predicted probability of response for each target. Then you employ a cutoff value for the probability of response and classify the predicted responses accordingly ○ Ex: with cutoff value of .5 - you would target a customer if predicted probability of response is greater than .5 otherwise you would not target the customer ● Stepwise regression ○ Step function applying both forward selection and backward selection, to select the best calibrated logistic model with the lowest AIC. If the name of your model fit from your initial model is bank.model.fit To explain effects of variables example: have to mention the comparison group - baseline for categorical variables Relative to people with white collar jobs, people with blue collar jobs are 0.75x as likely (or 25% less likely) to respond “yes” to the bank offer. Relative to people with white collar jobs, people not in the workforce are 1.76x as likely (or 76% more likely) to respond “yes” to the bank offer. Relative to people with primary education only, people with secondary education are 0.94x as likely (or 6% less likely) to respond yes to the offer. …Etc.

Interpretation for numeric variables were in chap 2 examples ● Look at all the assessment tools: AIC, McFadden’s, analysis of deviance table Validation tests ● Density lattice plots ● Cutoff value and predicted probabilities are compared to actual outcomes in confusion matrix ● Cohen’s kappa New validation test ● ROC Curve ○ Does not rely on predetermined cutoff value ○ Area under the receiver operation characteristic (ROC) curve is preferred index of classification performance, especially for low base rate problems (...