Framingham heart study logistic regression model

framingham.R

library(caTools)
library(ROCR)

framingham <- read.csv('framingham.csv')

set.seed(1000)

split <- sample.split(framingham$TenYearCHD, SplitRatio = 0.65)
train <- subset(framingham, split == TRUE)
test <- subset(framingham, split == FALSE)

framinghamLog <- glm(TenYearCHD ~ ., data=train, family=binomial)

predictTest <- predict(framinghamLog, type='response', newdata=test)
table(test$TenYearCHD, predictTest > 0.5)
# FALSE TRUE
# 0  1069    6
# 1   187   11

# Accuracy: 0.848
(1069+11) / (1069 + 6 + 187 + 11)

# Baseline accuracy, always predict 0 (no CHD): 0.844
(1069+6)/(1069+6+187+11)

ROCRpred <- prediction(predictTest, test$TenYearCHD)
as.numeric(performance(ROCRpred, 'auc')@y.values)
# AUC on test set: 0.742

# Sensitivity (true positive rate) = TP / (TP+FN): 0.05555556
11 / (11 + 187)

# Specificity (true negative rate) = TN / (TN+FP): 0.9944186
1069 / (1069 + 6)