The Bank Marketing Data includes actual information for a direct marketing effort by a bank. We will attempt to construct a model with just a few of the available attributes. We are interested in classifying whether a customer will respond positively to the marketing effort offering a term deposit. 

The attributes you are to use are age, balance, duration, campaign, pdays, and previous; explanations of these numeric variables are in the file. After your initial analysis you may wish to transform some of the remaining variables to attempt to estimate a more complete model.

Use a Logit model for the estimate making sure to request an “analysis of coefficients” in XLMiner. As usual, use a 60/40 split for the training and validation data sets and request a full set of lift charts.

Does this estimate using only some of the available attributes do better than a naïve model in the overall misclassification rate? What if you examine the lift chart? Recall that the lift chart reorders the data from most likely to accept a marketing offer to least likely to accept such an offer. Now does the algorithm appear to have explanatory power (i.e., could you successfully use it to suggest who to market to in the first place)?

Which of the attributes that you selected appear to have the greatest effect on the classification? How certain are you that these attributeshave an effect on the classification?

By creating dummy variables and categorical variables for the attributes that you did not use in this exercise already you may extend the analysis in order to refine the algorithm. Evaluate the resulting output in the same manner described above and compare the two outputs. Did the addition of the extra attributes to the Logit model add additional explanatory power?