Exercise 3.R

# ------------------------------------------------------------------
# EXERCISE 3
# Use the birthwt data in the MASS package to construct a model for low birth 
# weight. Are there any features which should be excluded from the model?
# ------------------------------------------------------------------

library(MASS)
library(caret)

birthwt2 <- birthwt #saving dataset in case I screw the original one up
head(birthwt)
str(birthwt)


## in order for confusionMatrix() to work,we actually need the "low"
## variable to be a factor. so let's do that before we split the data
## and start training.

birthwt$low <- factor(birthwt$low, labels = c("High","Low"))

# now split + train
index = createDataPartition(birthwt$low, list = FALSE, p = 0.8)[,1]
birthwt.train = birthwt[index,]
birthwt.test = birthwt[-index,]


birthwt.glm <- train(low ~ ., data = birthwt.train, method = "glm", family = binomial())
#no variable is significant?

confusionMatrix(predict(birthwt.glm, birthwt.test), birthwt.test$low, positive = "Low") #100% accuracy, wow.