University of Southern California BUAD 425 -Trojan Hose QUESTIONS WITH VERIFIED SOLUTIONS-GRADED A+

Document Content and Description Below

Trojan  Horse:  Unexpected  Style  at  Your  Door1 Case  Questions:   Decision  Tree 1. Using  the  training  data,  build  a  decision ...  tree  model  to  predict  probability  that   a  member  will  purchase  a  box.    If  you  set  a  cut-­‐off  threshold  of  .15,  how  many   members  would  you  target in  the  training  set?    Which  ones?     There  are  3  leaves  we  target  with  36  +  107  +  189  =  332  members  in  the   training  set.    The  leaves  are: • R  <  16  &&  ClassicGentleman  >=  1  &&  Rugged  >=2 • R  <  16  &&  ClassicGentleman  >=  1  &&  Rugged  <2 • R  <  16  &&  ClassicGentleman  <  1  &&  R  <  8 We  can  simplify  the  description  of  these  members  to: • R<  16  and  ClassicGentleman  >=  1 • R  <  8  and  ClassicGentleman  <  1 2. What  is  the  R2 of  your  model?    In  simple  words,  what  does  this  mean? The  R2  is  10.5%  on  the  training  data,  7.9%  on  the  testing  data.    In  words,  this   means  that  the  model  only  explains  about  10%  of  the  variability  in  the  data,   which  doesn’t  seem  like  very  much.    We’ll  see  later,  though,  that  it  still  has   enough  classification  accuracy  to  improve  our  marketing  campaign.     3. Examine  each  of  the  split  variables.    Do  they  make  intuitive/business  sense?     Explain  each  one. • Splitting  on  R  is  intuitive.    Customers  who  haven’t  purchased  in  a  long   time  probably  don’t  like  the  service  and  won’t  purchase  again. • Splitting  on  ClassicGentleman  probably  occurs  because  members  who   like  “ClassicGentleman” items like  the  current  box.     • The  box  is  also  probably  more  on  the  “Rugged”  side,  explaining  why   customers  who  purchase  from  Rugged  are  more  likely  to  buy.     • The  final  split  is  somewhat  more  confusing.    Customers  who  haven’t   bought  a  ClassicGentleman  product,  but  have  purchased  something   within  the  last  8  months  do  tend  to  like  the  box.    This  split  merits   further  investigation.    One  possibility,  e.g.,  is  that  members  in  this  box   are  new  to  the  service  and,  hence  haven’t  had  the  chance  to  buy  a   ClassicGentleman  product,  but  would  if  they were  offered  one.    If  they   had,  they’d  be  part  of  the  other  leaves.   These  splits  also  suggest  that  the  current  box  is  very  “ClassicGentleman”,  but  has   “Rugged”  elements  to  it.    Importantly,  this  description  of  the  box  was  generated [Show More]

Last updated: 1 year ago

Preview 1 out of 6 pages