Re: 문제197. (오늘의 마지막 문제)미국 대학 입학 데이터인 binary.csv 를 가지고 서포트 벡터 머신 모델을

[19기_김진경]

# 데이터 로드 
binary<-read.csv('binary.csv', stringsAsFactors = T) 
head(binary)
str(binary)

# admit 변수를 팩터형으로 변환
binary$admit <- as.factor(binary$admit)
binary$rank<-as.numeric(binary$rank)
binary$gre<-as.numeric(binary$gre)

str(binary)

# 훈련 데이터와 테스트 데이터를 분리
library(caret)
set.seed(1)

train_num <- createDataPartition(binary$admit, p = 0.8, list = FALSE)
train_data <- binary[train_num,]
test_data <- binary[-train_num,]

nrow(train_data) # 321
nrow(test_data)  # 79

normalize <- function(x) {
  return((x-min(x))/(max(x)-min(x)))
}

train_data_n<-as.data.frame(lapply(train_data[ , -1], normalize))
train_data2<-cbind(train_data[ , c('admit')], train_data_n)
names(train_data2)[1]<-'admit'
train_data2
test_data_n<-as.data.frame(lapply(test_data[ , -1], normalize))
test_data2<-cbind(test_data[ , c('admit')], test_data_n)
names(test_data2)[1]<-'admit'
test_data2

summary(test_data2)


# 모델 생성 및 하이퍼파라미터 설정
library(e1071)

# C와 gamma 값을 조정하여 모델 훈련
set.seed(1)

C_values <- 2^(-10:10)
gamma_values <- 2^(-10:10)

results <- data.frame( C=numeric(), gamma=numeric(), accuracy=numeric() )

for ( C in  C_values) {
  for ( gamma in gamma_values) { 
    set.seed(1)
    binary_svm_model <- svm(admit ~ . , data = train_data2, 
                            kernel = "radial", cost = C, gamma = gamma )
    result <- predict(binary_svm_model, test_data2[, -1])   
    result
    
    accuracy <- sum(result == test_data2$admit) / length(test_data2$admit)
    results <- rbind( results, data.frame(C=C, gamma=gamma, accuracy=accuracy))
  }
}
results

train_data2

library(doBy)
orderBy(~ -accuracy, results)

> orderBy(~ -accuracy, results)
                  C           gamma  accuracy
197    0.5000000000    0.1250000000 0.7215190
328   32.0000000000    4.0000000000 0.7215190
198    0.5000000000    0.2500000000 0.7088608
199    0.5000000000    0.5000000000 0.7088608
217    1.0000000000    0.0625000000 0.7088608
219    1.0000000000    0.2500000000 0.7088608
225    1.0000000000   16.0000000000 0.7088608
226    1.0000000000   32.0000000000 0.7088608
240    2.0000000000    0.2500000000 0.7088608
258    4.0000000000    0.0312500000 0.7088608
260    4.0000000000    0.1250000000 0.7088608