# Graphical representation of cross-validation errors for regression

What are some good ways of presenting/comparing cross-validated RMSE errors for regression using various models, graphically via plots? As of now, I have been presenting the quantitative results in tabular form.

-

If just a series of RMSE, ideally a bar chart should do if you don't have extra information to add. However, if you lined up the RMSE in a nice descending order in the table, I don't think a plot is necessary.

If you wish to emphasize the change in residual for individual point, a parallel coordinate plot should work quite well. See a pretty rough example below.

attach(iris)

m01 <- lm(Sepal.Length~Sepal.Width)
m02 <- lm(Sepal.Length~Sepal.Width + Petal.Length)
m03 <- lm(Sepal.Length~Sepal.Width + Petal.Length + as.factor(Species))

res1 <- m01$res res2 <- m02$res
res3 <- m03\$res

group <- c(rep(1,150), rep(2,150), rep(3,150))
allres <- c(res1, res2, res3)

plot(group, allres, pch=16, col="#ff000030", axes=F, ylab="Residual", xlab="")
axis(side=1, label=c("Model 1", "Model 2", "Model 3"), at=c(1,2,3))
axis(side=2)
abline(h=0)

for (i in seq(1, 150)){
lines(c(group[i], group[i+150], group[i+300]),
c(allres[i], allres[i+150], allres[i+300]), col="#ff000030")
}

-