Shapley value regression for brand driver analysis

[md2019]

Hi all,

Trying my hands on shapley regression model for brand driver importance. I have a quick question (pretty basic might be as well). I have multi-coded dependent variable. So the dependent variable question is " which brands would be you consider buying?" and then the answer could be X or X,Y or X,Y,Z (X.Y, Z are the names of brands). How shall I transform this one? Shall I take dummy categorical variable like X=1, Y=2, Z=3 or just consider the first preference among the ones mentioned by the customers? Can anyone help? Many thanks.

[nirgrahamuk]

Is there any limit to how many brands can be in the list?
I'd probably one hot encode it

[md2019]

Thanks, Yeah..there is a limit which is 3 at max

[nirgrahamuk]

library(tidyverse)

( ex1 <- tibble(
  id = 1:7,
   brandsliked =c("x,y,z","x,y","z","y,z","a","a,x","b,y")))
# # A tibble: 7 x 2
# id brandsliked
# <int> <chr>      
#   1    x,y,z      
#   2    x,y        
#   3    z          
#   4    y,z        
#   5    a          
#   6    a,x        
#   7    b,y   

(ex2 <- ex1 %>% separate(col = brandsliked,into=paste0("brnd",1:3),sep = ","))

(ex3 <- pivot_longer(ex2,
                     cols=c(brnd1:brnd3)))

library(caret)
ex4 <- dummyVars(" ~ value + id", data = ex3)
(ex5 <- data.frame(predict(ex4, newdata = ex3)))

(ex6 <- group_by(ex5,
                 id) %>% summarise(across(starts_with("value"),max,na.rm=TRUE)))
# # A tibble: 7 x 6
# id   valuea valueb valuex valuey valuez
# <dbl>  <dbl>  <dbl>  <dbl>  <dbl>  <dbl>
#   1      0      0      1      1      1
#   2      0      0      1      1      0
#   3      0      0      0      0      1
#   4      0      0      0      1      1
#   5      1      0      0      0      0
#   6      1      0      1      0      0
#   7      0      1      0      1      0