spam_or_ham.R

# spam or ham
# load library
library(kernlab)
library(naivebayes)
library(dplyr)

# load dataset into RStudio
data("spam")

# tibble dataframe
spam <- as.tbl(spam)
glimpse(spam) # 58 variables, 4601 records

# type variable
table(spam$type)
table(spam$type) / nrow(spam) # imbalanced classes

# any missing value
mean(complete.cases(spam)) # no missing value

# split dataset
set.seed(99)
id <- sample(nrow(spam), .8*nrow(spam))
train_set <- spam[id, ]; nrow(train_set)
test_set <- spam[-id, ]; nrow(test_set)

# train model
nb_model <- naive_bayes(type ~ ., data = train_set, laplace = 2)

# create a summary function
nb_summary <- function(nb_model, train_set, test_set){
   
   train_acc <- mean(predict(nb_model) == train_set$type)
   test_acc <- mean(predict(nb_model, newdata = test_set) == test_set$type)
   
   cat("===== Model Summary =====\n")
   cat("train accuracy: ", round(train_acc*100, 2), "%", sep = "")
   cat("\n")
   cat(" test accuracy: ", round(test_acc*100, 2), "%", sep = "")
}

# see the accuracy of train and test data
nb_summary(nb_model, train_set, test_set)