如何使用r？

Question

set.seed(123)
data <- twoClassSim(1000)
trainIndex <- createDataPartition(data$Class, p = 0.8, list = FALSE)
trainData <- data[trainIndex, ]
testData <- data[-trainIndex, ]

# Train a logistic regression model
model <- glm(Class ~ ., data = trainData, family = binomial)

# Get predicted probabilities
test_probs <- predict(model, testData, type = "response")

# Compute AUC BEFORE calibration
auc_before <- pROC::roc(testData$Class, test_probs)$auc

# Fit an isotonic regression model
iso_model <- isoreg(test_probs, as.numeric(testData$Class) - 1)

# Get calibrated probabilities from isotonic regression
iso_calibrated_probs <- approx(iso_model$x, iso_model$yf, xout = test_probs)$y

# Compute AUC AFTER calibration
auc_after <- pROC::roc(testData$Class, iso_calibrated_probs)$auc

auc_befor导致0.9116，但AUC_AFTER仅为0.52。我相信等渗回归的合适概率应该是有问题的。有人可以指出我的代码中的错误，以便我获得相同的AUC吗？非常感谢它
    

我不确定您正在使用

approx

函数。存档的

Rfutilities

软件包提供了

Answer 1

函数，该功能似乎对您的需求有用。

library(rfUtilities)

iso_calibrated_probs <- probability.calibration(as.numeric(testData$Class)-1, test_probs, regularization = TRUE)

# Compute AUC AFTER calibration
auc_after <- pROC::roc(testData$Class, iso_calibrated_probs)$auc
auc_after 
Area under the curve: 0.9119

plot(density(test_probs), col="red", xlim=c(0,1), 
     ylab="Density", xlab="probabilities",
     main="Calibrated probabilities", las=1)
lines(density(iso_calibrated_probs), col="blue")
legend("topright", legend=c("original", "calibrated"), 
       lty = c(1,1), col=c("red","blue"))