review

review

library(dplyr)
library(rsample)
library(caret)
library(partykit)
library(e1071)

read data diabetes

diabetes <- read.csv("data_input/diabetes.csv")
head(diabetes)

cross validation

set.seed(100)
idx <- initial_split(diabetes, prop = 0.8, strata = "diabetes")
diabetes_train <- training(idx)
diabetes_test <- testing(idx)
prop.table(table(diabetes_train$diabetes))

## 
##      neg      pos 
## 0.650974 0.349026

prop.table(table(diabetes_test$diabetes))

## 
##       neg       pos 
## 0.6513158 0.3486842

prop.table(table(diabetes$diabetes))

## 
##       neg       pos 
## 0.6510417 0.3489583

modeling

1. naive bayes

• pada naive bayes semua predictor memiliki bobot yang sama (tidak ada yang 1 variabel yg lebih penting dibanding yang lain)

diabetes_bayes <- naiveBayes(formula = diabetes~., data = diabetes_train, laplace = 0)

predict

# 2 jenis type : raw, class
bayes_pred <- predict(diabetes_bayes, diabetes_test, type = "class")
bayes_pred_prob <- predict(diabetes_bayes, diabetes_test, type = "raw")

confusion matrix

confusionMatrix(bayes_pred, diabetes_test$diabetes, positive = "pos")[[3]][1]

## Accuracy 
##     0.75

2. decision tree

diabetes_tree <- ctree(formula = diabetes~., data = diabetes_train)
plot(diabetes_tree, type = "simple")

predict

tree_pred <- predict(diabetes_tree, diabetes_test)
tree_pred_prob <- predict(diabetes_tree, diabetes_test, type = "prob")

confusionMatrix(tree_pred, diabetes_test$diabetes, positive = "pos")[[3]][1]

##  Accuracy 
## 0.6710526

3. random forest

ctrl <- trainControl(method = "repeatedcv", number = 5, repeats = 2)
diabetes_forest <- train(diabetes~., data = diabetes_train, method = "rf", ntree = 200, trControl = ctrl)
plot(diabetes_forest)

varImp(diabetes_forest)

## rf variable importance
## 
##          Overall
## glucose  100.000
## mass      49.388
## pedigree  29.281
## age       21.953
## pressure  11.606
## pregnant   7.495
## triceps    2.478
## insulin    0.000

plot(diabetes_forest$finalModel)
legend("topright", colnames(diabetes_forest$finalModel$err.rate), fill = 1:3)

forest_pred <- predict(diabetes_forest, diabetes_test)

confusionMatrix(forest_pred, diabetes_test$diabetes, positive = "pos")[[3]][1]

##  Accuracy 
## 0.7631579

library(ROCR)

## Loading required package: gplots

## 
## Attaching package: 'gplots'

## The following object is masked from 'package:stats':
## 
##     lowess

pred <- prediction(bayes_pred_prob[,2], diabetes_test$diabetes)
perf <- performance(pred, "tpr", "fpr")
plot(perf)

auc <- performance(pred,"auc")
auc@y.values[[1]]

## [1] 0.8151325