How to vote binary outcome of predictions

Question

Considering Titanic data set, I created several predictions on the survival and I want to create the final Survival based on a vote system, meaning that if the majority of the predictions stipulate that the passenger survived the final outcome is 1, 0 otherwise

> str(temp)
'data.frame':   179 obs. of  3 variables:
 $ predictions_ldm    : Factor w/ 2 levels "0","1": 1 1 1 1 1 1 1 1 1 1 ...
 $ predictions_qda    : Factor w/ 2 levels "0","1": 1 1 1 1 2 1 1 1 1 1 ...
 $ predictions_glm_age: Factor w/ 2 levels "0","1": 1 1 1 1 1 1 1 1 1 1 ...

> temp[c(4,5,12),]
   predictions_ldm predictions_qda predictions_glm_age
4                0               0                   0
5                0               1                   0
12               1               1                   0

I want the result to be

> temp[c(4,5,12),]
   predictions_ldm predictions_qda predictions_glm_age            Survived
4                0               0                   0                   0
5                0               1                   0                   0
12               1               1                   0                   1

How can I achieve this?

you could get the rowMeans, then if they are greater than 0.5 put 1, else 0? like `ifelse(rowMeans(temp) > 0.5, 1, 0)`? — morgan121, Jun 03 '20 at 21:42
I found a solution, but I don't think that is the optimal one `votedSurvival <- as.factor(as.numeric(apply(temp,1,FUN = function(z) {mean(as.numeric(z))>0.5})))` — beasst, Jun 03 '20 at 21:51
good solution, another similar way is to use `apply(temp, 1, function(x) names(sort(-table(x[1]))))` in the function, which doesn't convert to numeric — morgan121, Jun 03 '20 at 21:58

score 2 · Answer 1 · answered Jun 03 '20 at 21:51

2

It's an unneccessary complicated solution using dplyr but I really wanted to use c_across(). At first I needed to convert your factors into integers keeping the 0-1 values.

temp %>%
  mutate(across(where(is.factor), function(x) { x %>% 
      as.character() %>% 
      as.integer()
    } )) %>%
  rowwise %>%
  mutate(Survived = c_across() %>% 
           mean() %>%
           round() %>%
           as.integer)

answered Jun 03 '20 at 21:51

Martin Gal

16,640
5
21
39

Thank you, i will try, I found a pretty complicated solution also `votedSurvival <- as.factor(as.numeric(apply(temp,1,FUN = function(z) {mean(as.numeric(z))>0.5})))` – beasst Jun 03 '20 at 21:53
Do you need `Survived` to be a factor? – Martin Gal Jun 03 '20 at 21:54
Yes. I want to get the accuracy `print(mean(votedSurvival == test_set$Survived))` – beasst Jun 03 '20 at 22:04

score 0 · Answer 2 · answered Jun 04 '20 at 05:17

You can use the Mode function defined here :

Mode <- function(x) {
  ux <- unique(x)
  ux[which.max(tabulate(match(x, ux)))]
}

and apply this function row-wise :

temp$Survived <- apply(temp[c(4,5,12),], 1, Mode)

Mode returns the most frequently occurring value from a vector.

How to vote binary outcome of predictions

2 Answers2