Boxplots with the statistical (wilcoxon) test

Question

I have to compare different ML algorithms using caret R package and then find the significance difference between each of these algorithms.

For instance, I am using my code as follows

nnet2 <- train(result ~ ., data = tr,
              method = "nnet",
               tuneLength = 15,
              metric = "MAE",
              preProc = c("center", "scale", "nzv"),
              trControl = ctrl)

getTrainPerf(nnet2)

svm2 <- train(result ~ ., data = tr,
             method = "svmRadial",

             tuneLength = 15,
             metric = "MAE",
             preProc = c("center", "scale", "nzv"),
             trControl = ctrl)

getTrainPerf(svm2)

and few other algorithms like that. Then I have performed wilcoxon test

wilcox.test (nnet2$resample$MAE, svm2$resample$MAE, paired=T)

My question is how can we put the results of the wilcoxon tests as a boxplot in R language?

Thanks

Answer 1

Using an example dataset:

library(caret)
library(mlbench)
library(ggpubr)
data(BostonHousing)
tr = BostonHousing

ctrl = trainControl(method="cv",number=10)

nnet2 <- train(medv ~ ., data = tr,
              method = "nnet",
               tuneLength = 5,
              metric = "MAE",
              preProc = c("center", "scale", "nzv"),
              trControl = ctrl)

svm2 <- train(medv ~ ., data = tr,
             method = "svmRadial",

             tuneLength = 5,
             metric = "MAE",
             preProc = c("center", "scale", "nzv"),
             trControl = ctrl)

Better to create a data.frame, with the MAE and a vector representing the model:

df = data.frame(MAE=c(nnet2$resample$MAE,svm2$resample$MAE),
model=rep(c("nnet","svm"),each=length(svm2$resample$MAE)))

ggboxplot(df, x = "model",y= "MAE",col="model",palette = c("#00AFBB", "#E7B800")) + 
stat_compare_means()