Peer-graded Assignment Part 5 Modeling Credit Card Default Risk and Customer Profitability.txt

1 a. My predictive model is: -(( Standardized Income + Standardized Credit Card Debt + Standardized Automobile Debt � 1! " (Standardized A#e + Standardized $ears at %mployer Applyin# t&e model: t&e Area 'nder t&e Curve (A'C is .). t&e Minimum Cost *er %vent is ,1. and is at t&res&old -1./. b. 0eer to t&e te2t bo2 belo3 or a response to part b o t&e 4uestion.

/ 'sin# t&e predictive model5 t&e applicant listed in t&e 4uestion 3ould &ave t&e ollo3in# score: -(( Standardized Income + Standardized Credit Card Debt + Standardized Automobile Debt � 1! " (Standardized A#e + Standardized $ears at %mployer -((-!.), + .16 + -!.!7

�

1! " (-!.!7 + ./) 8 /.!!

Since /.!! is &i#&er t&an t&e t&res&old o -1./ (previously stated5 t&e applicant 3ould be included in t&e selection (3ould be approved.

) 'sin# t&e predictive model on t&e 9rainin# Set5 t&e avera#e proit per applicant is: 161(9rue *ositive Count " 65!!! (Avera#e *roits per *roitable Customer 8 765!!! +  (alse ;e#ative Count " -65!! (Avera#e
6 'sin# t&e predictive model (compared to no model5 t&e incremental inancial value on t&e 9rainin# Set is: 65!! (Cost per alse ;e#ative " /= (proportion o unproitable applicants 8 15// ,1. (*redictive Model Minimum Cost per %vent 8 ))).

 %valuate your model on t&e 9est Set data. >o3 conident are you t&at your model does not over-it t&e 9rainin# Set data? 9&e only basis to evaluate over-ittin# is to #ive t&e same metrics on t&e 9est Set and 9rainin# Set5 and compare t&em. 0eer to t&e te2t bo2 belo3 or a response concernin# over-ittin#.

7 %valuate your model on t&e 9est Set data. >o3 conident are you t&at your model does not over-it t&e 9rainin# Set data? A. C&oose bet3een t&ree broad de#rees o conidence: � very� � some3&at� or � not at all.� (;ote t&at � not at all� is still an acceptable ans3er i you #ive persuasive reasons or 3&y you c&ose t&is ans3er. @. %2plain t&e evidence your de#ree o conidence is based upon. $our e2planation s&ould include t&e test set proits and trainin# set proits per applicant. >o3 muc& conidence to &ave in t&e model must relate to t&e relations&ip bet3een t&e proits-per-applicant on t&e 9rainin# Set and t&e 9est Set 'sin# t&e predictive model on t&e 9est Set5 t&e Area 'nder t&e Curve (A'C is .6. Additionally5 usin# t&e t&res&old o -1./ (rom t&e predictive model on t&e trainin# set5 t&e minimum cost per event 151/. 'sin# t&e test set data5 t&e avera#e proit per applicant is: 1)6 (9rue *ositive Count " 65!!! (Avera#e *roits per *roitable Customer 8 )75!!! + 17 (alse ;e#ative Count " -65!! (Avera#e