The president of the bank asks you to design and implement a predictive model to determine which future applicants should be approved for a credit card and which rejected.
your new goal is to develop a binary classification process for "approve" or "reject" designed to maximize total bank profits. (put another way, you want to maximize average profits per applicant, including in the total number of applicants those that are rejected).
assignment details:
assume you have available to you the six standardized data inputs only (no eggertopia scores). design your model on the training set only.
you may adapt your own previous binary classification model for default, the example binary classification model for default given in the learning points for quiz 1 and 2,
you may also use the linear regression model for profitability, or create a new model for this purpose.
no methods outside the scope of what has already been taught in this course need to be, or should be, used.
note an important change in the bank’s assumptions:
the bank has learned that not all defaulters are unprofitable and not all non-defaulters are profitable. therefore you should no longer use the bank’s "old" estimates for rate of default (25%), average losses per defaulter (-$5,000) average profits per non-defaulter ($2,500), and average profits per applicant when using no model ($625).
instead, the new relevant profitability metrics assumptions the bank gives you to use are:
the proportion of applicants that are unprofitable (again 25%, but a different 25%), average losses per unprofitable customer (-$4,900), average profits per profitable customer ($4,000), and average profits per applicant when using no model ($1,775).
a "perfect" model that excluded all unprofitable and included all profitable customers would have an estimated profitability per applicant of ($4,000)*.75 = $3,000
directions:
you will submit a summary of your model and its performance to the bank president. the summary will address the following questions:
what is your predictive model?
a. describe the arithmetic clearly so that another learner could implement your model on new standardized input data if they wished.
b. give an example of the score you would assign the following applicant, whether they would be approved or rejected for a credit card and why.
age: -0.06
years at employer: 0.23
years at address: -0.58
income: -0.38
credit card debt: 0.14
auto debt: -0.06