performance (version 0.9.1)

performance_roc: Simple ROC curve

Description

This function calculates a simple ROC curves of x/y coordinates based on response and predictions of a binomial model.

Usage

performance_roc(x, ..., predictions, new_data)

Value

A data frame with three columns, the x/y-coordinate pairs for the ROC curve (Sensitivity and Specificity), and a column with the model name.

Arguments

x

A numeric vector, representing the outcome (0/1), or a model with binomial outcome.

...

One or more models with binomial outcome. In this case, new_data is ignored.

predictions

If x is numeric, a numeric vector of same length as x, representing the actual predicted values.

new_data

If x is a model, a data frame that is passed to predict() as newdata-argument. If NULL, the ROC for the full model is calculated.

Examples

library(bayestestR)
data(iris)

set.seed(123)
iris$y <- rbinom(nrow(iris), size = 1, .3)
folds <- sample(nrow(iris), size = nrow(iris) / 8, replace = FALSE)
test_data <- iris[folds, ]
train_data <- iris[-folds, ]

model <- glm(y ~ Sepal.Length + Sepal.Width, data = train_data, family = "binomial")
as.data.frame(performance_roc(model, new_data = test_data))

roc <- performance_roc(model, new_data = test_data)
area_under_curve(roc$Specificity, roc$Sensitivity)

m1 <- glm(y ~ Sepal.Length + Sepal.Width, data = iris, family = "binomial")
m2 <- glm(y ~ Sepal.Length + Petal.Width, data = iris, family = "binomial")
m3 <- glm(y ~ Sepal.Length + Species, data = iris, family = "binomial")
performance_roc(m1, m2, m3)

# if you have `see` package installed, you can also plot comparison of
# ROC curves for different models
if (require("see")) plot(performance_roc(m1, m2, m3))