Linear Regression.R

## correlation
cor(mtcars$mpg, mtcars$hp)
plot(mtcars$mpg, mtcars$hp)

plot(mtcars$mpg, mtcars$hp, pch = 16)

cor(mtcars[ , c("mpg", "wt", "hp")])
## dplyr
library(dplyr)

mtcars %>% 
  select(mpg, wt, hp) %>%
  cor()

cor.test(mtcars$hp, mtcars$mpg)
## linear Regression

lmfit <- lm(mpg ~ hp, data = mtcars)
summary(lmfit)
lmfit$coefficient
lmfit$coefficients[1] + lmfit$coefficients[2] *200
lmfit$coefficients[[1]] + lmfit$coefficients[[2]] *200
newcars<- data.frame(hp = c(450, 400, 250, 320))
## predict
predict(lmfit, newdata = newcars)
newcars$mpg_pred <- predict(lmfit, newdata = newcars)
newcars$hp_pred <- NULL
newcars
summary(mtcars$hp)

## Root mean squared error(rmse)
## mutiple linear regression
## mpg = f(hp, wt, am)

lmfitv2 <- lm(mpg ~ hp + wt + am, data = mtcars)
coefs <- coef(lmfitv2)
coefs[[1]] + coefs[[2]]*200 + coefs[[3]]*3.5 + coefs[[4]]*1

##build full Model
lmfull <- lm(mpg ~ ., data = mtcars)
mtcars$predicted <- predict(lmfull)
## RMSE
squared_error <- (mtcars$mpg - mtcars$predicted) ** 2 
(rmse <- sqrt(mean(squared_error)))

## split data 
## random split
set.seed(42)
sample(1:10, 3)
x <- nrow(mtcars)
id <- sample(1:x, size = x*0.8)
train_data <- mtcars[id, ]
test_data <- mtcars[id, ]

## train model
model1 <- lm(mpg ~ hp + wt + am + disp, data = train_data)
p_train <- predict(model1)
rmse <- sqrt(mean((train_data$mpg - p_train)**2))
## test model
p_test <- predict(model1, newdata = test_data)
error_test <- test_data$mpg - p_test
(rmse_test <- sqrt(mean((train_data$mpg - p_train)**2)))


##print result
cat("RMSE Train :", rmse_test, "\nRMSE test :", rmse_test)