library(tidyverse)
library(caret)
library(glmnet)
creditdata <- read_excel("R bestanden/creditdata.xlsx")
df <- as.data.frame(creditdata)
df <- na.omit(df)
df$married <- as.factor(df$married)
df$graduate_school <- as.factor(df$graduate_school)
df$high_school <- as.factor(df$high_school)
df$default_payment_next_month <- as.factor(df$default_payment_next_month)
df$sex <- as.factor(df$sex)
df$single <- as.factor(df$single)
df$university <- as.factor(df$university)
set.seed(123)
training.samples <- df$default_payment_next_month %>%
createDataPartition(p = 0.8, list = FALSE)
train.data <- df[training.samples, ]
test.data <- df[-training.samples, ]
x <- model.matrix(default_payment_next_month~., train.data)[,-1]
y <- ifelse(train.data$default_payment_next_month == 1, 1, 0)
cv.lasso <- cv.glmnet(x, y, alpha = 1, family = "binomial")
lasso.model <- glmnet(x, y, alpha = 1, family = "binomial",
lambda = cv.lasso$lambda.1se)
x.test <- model.matrix(default_payment_next_month ~., test.data)[,-1]
probabilities <- lasso.model %>% predict(newx = x.test)
predicted.classes <- ifelse(probabilities > 0.5, "1", "0")
observed.classes <- test.data$default_payment_next_month
mean(predicted.classes == observed.classes)
Hi guys,
I'm new in R and I've been trying to use the exact code as on this website http://www.sthda.com/english/articles/36-classification-methods-essentials/149-penalized-logistic-regression-essentials-in-r-ridge-lasso-and-elastic-net/ to perform a logistic ridge regression. My aim is to predict if a client has credit card default or not, and we have a data set with factor variables as well as numerical variables. The problem is that most of my probabilities are negative and smaller than -1, so -2.6, -1.4 etc. Does anyone know what is going wrong here?
Thanks in advance for the help!
Just like for glm, by default the predict function for glmnet returns predictions on the scale of the link function, which aren't probabilities.
To get the predicted probabilities, add type = "response" to the predict call:
probabilities <- lasso.model %>% predict(newx = x.test, type = "response")