Description for Support Vector Machine with Radial Basis Kernel using Package kernlab
Support Vector Machine의 장점
Support Vector Machine의 단점
실습 자료 : 유니버셜 은행의 고객 2,500명에 대한 자료(출처 : Data Mining for Business Intelligence, Shmueli et al. 2010)이며, 총 13개의 변수를 포함하고 있다. 이 자료에서 Target은
Personal Loan
이다.
pacman::p_load("data.table", "dplyr",
"caret",
"ggplot2", "GGally",
"kernlab")
UB <- fread("../Universal Bank_Main.csv") # 데이터 불러오기
UB %>%
as_tibble
# A tibble: 2,500 × 14
ID Age Experience Income `ZIP Code` Family CCAvg Education
<int> <int> <int> <int> <int> <int> <dbl> <int>
1 1 25 1 49 91107 4 1.6 1
2 2 45 19 34 90089 3 1.5 1
3 3 39 15 11 94720 1 1 1
4 4 35 9 100 94112 1 2.7 2
5 5 35 8 45 91330 4 1 2
6 6 37 13 29 92121 4 0.4 2
7 7 53 27 72 91711 2 1.5 2
8 8 50 24 22 93943 1 0.3 3
9 9 35 10 81 90089 3 0.6 2
10 10 34 9 180 93023 1 8.9 3
# ℹ 2,490 more rows
# ℹ 6 more variables: Mortgage <int>, `Personal Loan` <int>,
# `Securities Account` <int>, `CD Account` <int>, Online <int>,
# CreditCard <int>
UB %<>%
data.frame() %>% # Data Frame 형태로 변환
mutate(Personal.Loan = ifelse(Personal.Loan == 1, "yes", "no")) %>% # Target을 문자형 변수로 변환
select(-1) # ID 변수 제거
# 1. Convert to Factor
fac.col <- c("Family", "Education", "Securities.Account",
"CD.Account", "Online", "CreditCard",
# Target
"Personal.Loan")
UB <- UB %>%
mutate_at(fac.col, as.factor) # 범주형으로 변환
glimpse(UB) # 데이터 구조 확인
Rows: 2,500
Columns: 13
$ Age <int> 25, 45, 39, 35, 35, 37, 53, 50, 35, 34, 6…
$ Experience <int> 1, 19, 15, 9, 8, 13, 27, 24, 10, 9, 39, 5…
$ Income <int> 49, 34, 11, 100, 45, 29, 72, 22, 81, 180,…
$ ZIP.Code <int> 91107, 90089, 94720, 94112, 91330, 92121,…
$ Family <fct> 4, 3, 1, 1, 4, 4, 2, 1, 3, 1, 4, 3, 2, 4,…
$ CCAvg <dbl> 1.6, 1.5, 1.0, 2.7, 1.0, 0.4, 1.5, 0.3, 0…
$ Education <fct> 1, 1, 1, 2, 2, 2, 2, 3, 2, 3, 3, 2, 3, 2,…
$ Mortgage <int> 0, 0, 0, 0, 0, 155, 0, 0, 104, 0, 0, 0, 0…
$ Personal.Loan <fct> no, no, no, no, no, no, no, no, no, yes, …
$ Securities.Account <fct> 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0,…
$ CD.Account <fct> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,…
$ Online <fct> 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 1, 0, 1,…
$ CreditCard <fct> 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0,…
# 2. Convert One-hot Encoding for 범주형 예측 변수
dummies <- dummyVars(formula = ~ ., # formula : ~ 예측 변수 / "." : data에 포함된 모든 변수를 의미
data = UB[,-9], # Dataset including Only 예측 변수 -> Target 제외
fullRank = FALSE) # fullRank = TRUE : Dummy Variable, fullRank = FALSE : One-hot Encoding
UB.Var <- predict(dummies, newdata = UB) %>% # 범주형 예측 변수에 대한 One-hot Encoding 변환
data.frame() # Data Frame 형태로 변환
glimpse(UB.Var) # 데이터 구조 확인
Rows: 2,500
Columns: 21
$ Age <dbl> 25, 45, 39, 35, 35, 37, 53, 50, 35, 34,…
$ Experience <dbl> 1, 19, 15, 9, 8, 13, 27, 24, 10, 9, 39,…
$ Income <dbl> 49, 34, 11, 100, 45, 29, 72, 22, 81, 18…
$ ZIP.Code <dbl> 91107, 90089, 94720, 94112, 91330, 9212…
$ Family.1 <dbl> 0, 0, 1, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, …
$ Family.2 <dbl> 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, …
$ Family.3 <dbl> 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, …
$ Family.4 <dbl> 1, 0, 0, 0, 1, 1, 0, 0, 0, 0, 1, 0, 0, …
$ CCAvg <dbl> 1.6, 1.5, 1.0, 2.7, 1.0, 0.4, 1.5, 0.3,…
$ Education.1 <dbl> 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
$ Education.2 <dbl> 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 0, 1, 0, …
$ Education.3 <dbl> 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1, 0, 1, …
$ Mortgage <dbl> 0, 0, 0, 0, 0, 155, 0, 0, 104, 0, 0, 0,…
$ Securities.Account.0 <dbl> 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, …
$ Securities.Account.1 <dbl> 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, …
$ CD.Account.0 <dbl> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
$ CD.Account.1 <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
$ Online.0 <dbl> 1, 1, 1, 1, 1, 0, 0, 1, 0, 1, 1, 0, 1, …
$ Online.1 <dbl> 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 1, 0, …
$ CreditCard.0 <dbl> 1, 1, 1, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1, …
$ CreditCard.1 <dbl> 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, …
# 3. Combine Target with 변환된 예측 변수
UB.df <- data.frame(Personal.Loan = UB$Personal.Loan,
UB.Var)
UB.df %>%
as_tibble
# A tibble: 2,500 × 22
Personal.Loan Age Experience Income ZIP.Code Family.1 Family.2
<fct> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
1 no 25 1 49 91107 0 0
2 no 45 19 34 90089 0 0
3 no 39 15 11 94720 1 0
4 no 35 9 100 94112 1 0
5 no 35 8 45 91330 0 0
6 no 37 13 29 92121 0 0
7 no 53 27 72 91711 0 1
8 no 50 24 22 93943 1 0
9 no 35 10 81 90089 0 0
10 yes 34 9 180 93023 1 0
# ℹ 2,490 more rows
# ℹ 15 more variables: Family.3 <dbl>, Family.4 <dbl>, CCAvg <dbl>,
# Education.1 <dbl>, Education.2 <dbl>, Education.3 <dbl>,
# Mortgage <dbl>, Securities.Account.0 <dbl>,
# Securities.Account.1 <dbl>, CD.Account.0 <dbl>,
# CD.Account.1 <dbl>, Online.0 <dbl>, Online.1 <dbl>,
# CreditCard.0 <dbl>, CreditCard.1 <dbl>
glimpse(UB.df) # 데이터 구조 확인
Rows: 2,500
Columns: 22
$ Personal.Loan <fct> no, no, no, no, no, no, no, no, no, yes…
$ Age <dbl> 25, 45, 39, 35, 35, 37, 53, 50, 35, 34,…
$ Experience <dbl> 1, 19, 15, 9, 8, 13, 27, 24, 10, 9, 39,…
$ Income <dbl> 49, 34, 11, 100, 45, 29, 72, 22, 81, 18…
$ ZIP.Code <dbl> 91107, 90089, 94720, 94112, 91330, 9212…
$ Family.1 <dbl> 0, 0, 1, 1, 0, 0, 0, 1, 0, 1, 0, 0, 0, …
$ Family.2 <dbl> 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, …
$ Family.3 <dbl> 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, …
$ Family.4 <dbl> 1, 0, 0, 0, 1, 1, 0, 0, 0, 0, 1, 0, 0, …
$ CCAvg <dbl> 1.6, 1.5, 1.0, 2.7, 1.0, 0.4, 1.5, 0.3,…
$ Education.1 <dbl> 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
$ Education.2 <dbl> 0, 0, 0, 1, 1, 1, 1, 0, 1, 0, 0, 1, 0, …
$ Education.3 <dbl> 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1, 0, 1, …
$ Mortgage <dbl> 0, 0, 0, 0, 0, 155, 0, 0, 104, 0, 0, 0,…
$ Securities.Account.0 <dbl> 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, …
$ Securities.Account.1 <dbl> 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, …
$ CD.Account.0 <dbl> 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
$ CD.Account.1 <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
$ Online.0 <dbl> 1, 1, 1, 1, 1, 0, 0, 1, 0, 1, 1, 0, 1, …
$ Online.1 <dbl> 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 1, 0, …
$ CreditCard.0 <dbl> 1, 1, 1, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1, …
$ CreditCard.1 <dbl> 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, …
ggpairs(UB, # In 2-1
columns = c("Age", "Experience", "Income", # 수치형 예측 변수
"ZIP.Code", "CCAvg", "Mortgage"),
aes(colour = Personal.Loan)) + # Target의 범주에 따라 색깔을 다르게 표현
theme_bw()
ggpairs(UB, # In 2-1
columns = c("Age", "Experience", "Income", # 수치형 예측 변수
"ZIP.Code", "CCAvg", "Mortgage"),
aes(colour = Personal.Loan)) + # Target의 범주에 따라 색깔을 다르게 표현
scale_colour_manual(values = c("#00798c", "#d1495b")) + # 특정 색깔 지정
scale_fill_manual(values = c("#00798c", "#d1495b")) + # 특정 색깔 지정
theme_bw()
ggpairs(UB, # In 2-1
columns = c("Age", "Income", # 수치형 예측 변수
"Family", "Education"), # 범주형 예측 변수
aes(colour = Personal.Loan, alpha = 0.8)) + # Target의 범주에 따라 색깔을 다르게 표현
scale_colour_manual(values = c("#E69F00", "#56B4E9")) + # 특정 색깔 지정
scale_fill_manual(values = c("#E69F00", "#56B4E9")) + # 특정 색깔 지정
theme_bw()
# Partition (Training Dataset : Test Dataset = 7:3)
y <- UB.df$Personal.Loan # Target
set.seed(200)
ind <- createDataPartition(y, p = 0.7, list = T) # Index를 이용하여 7:3으로 분할
UB.trd <- UB.df[ind$Resample1,] # Training Dataset
UB.ted <- UB.df[-ind$Resample1,] # Test Dataset
# Standardization
preProcValues <- preProcess(UB.trd,
method = c("center", "scale")) # Standardization 정의 -> Training Dataset에 대한 평균과 표준편차 계산
UB.trd <- predict(preProcValues, UB.trd) # Standardization for Training Dataset
UB.ted <- predict(preProcValues, UB.ted) # Standardization for Test Dataset
glimpse(UB.trd) # 데이터 구조 확인
Rows: 1,751
Columns: 22
$ Personal.Loan <fct> no, no, no, no, no, no, no, yes, no, no…
$ Age <dbl> -0.05431273, -0.57446728, -0.92123699, …
$ Experience <dbl> -0.12175295, -0.46882565, -0.98943471, …
$ Income <dbl> -0.85867297, -1.35649686, 0.56986515, -…
$ ZIP.Code <dbl> -1.75250883, 0.88354520, 0.53745994, -1…
$ Family.1 <dbl> -0.6355621, 1.5725118, 1.5725118, -0.63…
$ Family.2 <dbl> -0.5774051, -0.5774051, -0.5774051, -0.…
$ Family.3 <dbl> 2.0037210, -0.4987865, -0.4987865, -0.4…
$ Family.4 <dbl> -0.5967491, -0.5967491, -0.5967491, 1.6…
$ CCAvg <dbl> -0.25119120, -0.53150921, 0.42157204, -…
$ Education.1 <dbl> 1.1482386, 1.1482386, -0.8704018, -0.87…
$ Education.2 <dbl> -0.6196534, -0.6196534, 1.6128838, 1.61…
$ Education.3 <dbl> -0.6408777, -0.6408777, -0.6408777, -0.…
$ Mortgage <dbl> -0.5664192, -0.5664192, -0.5664192, -0.…
$ Securities.Account.0 <dbl> -2.7998134, 0.3569627, 0.3569627, 0.356…
$ Securities.Account.1 <dbl> 2.7998134, -0.3569627, -0.3569627, -0.3…
$ CD.Account.0 <dbl> 0.2613337, 0.2613337, 0.2613337, 0.2613…
$ CD.Account.1 <dbl> -0.2613337, -0.2613337, -0.2613337, -0.…
$ Online.0 <dbl> 1.2486195, 1.2486195, 1.2486195, 1.2486…
$ Online.1 <dbl> -1.2486195, -1.2486195, -1.2486195, -1.…
$ CreditCard.0 <dbl> 0.6408777, 0.6408777, 0.6408777, -1.559…
$ CreditCard.1 <dbl> -0.6408777, -0.6408777, -0.6408777, 1.5…
glimpse(UB.ted) # 데이터 구조 확인
Rows: 749
Columns: 22
$ Personal.Loan <fct> no, no, no, no, no, no, no, no, no, no,…
$ Age <dbl> -1.7881612, -0.7478521, 1.2460737, 0.81…
$ Experience <dbl> -1.68358012, -0.64236200, 0.83269699, 0…
$ Income <dbl> -0.53400522, -0.96689556, -1.11840718, …
$ ZIP.Code <dbl> -1.17304370, -0.59585545, 1.07366441, 0…
$ Family.1 <dbl> -0.6355621, -0.6355621, 1.5725118, 1.57…
$ Family.2 <dbl> -0.5774051, -0.5774051, -0.5774051, -0.…
$ Family.3 <dbl> -0.4987865, -0.4987865, -0.4987865, -0.…
$ Family.4 <dbl> 1.6747892, 1.6747892, -0.5967491, -0.59…
$ CCAvg <dbl> -0.19512759, -0.86789083, -0.25119120, …
$ Education.1 <dbl> 1.1482386, -0.8704018, -0.8704018, -0.8…
$ Education.2 <dbl> -0.6196534, 1.6128838, -0.6196534, 1.61…
$ Education.3 <dbl> -0.6408777, -0.6408777, 1.5594690, -0.6…
$ Mortgage <dbl> -0.5664192, 0.9609885, -0.5664192, -0.5…
$ Securities.Account.0 <dbl> -2.7998134, 0.3569627, 0.3569627, -2.79…
$ Securities.Account.1 <dbl> 2.7998134, -0.3569627, -0.3569627, 2.79…
$ CD.Account.0 <dbl> 0.2613337, 0.2613337, 0.2613337, 0.2613…
$ CD.Account.1 <dbl> -0.2613337, -0.2613337, -0.2613337, -0.…
$ Online.0 <dbl> 1.2486195, -0.8004271, -0.8004271, 1.24…
$ Online.1 <dbl> -1.2486195, 0.8004271, 0.8004271, -1.24…
$ CreditCard.0 <dbl> 0.6408777, 0.6408777, -1.5594690, -1.55…
$ CreditCard.1 <dbl> -0.6408777, -0.6408777, 1.5594690, 1.55…
Package "kernlab"
는 커널 기반의 기계 학습 알고리듬을 R에서 구현한 Package이며, 해당 Package를 이용하여 Support Vector Machine를 수행하면 Kernel 함수가 Radial Basis
일 때, 최적의 sigma (gamma)
값을 자동으로 찾아준다는 장점이 있다. 함수 ksvm()
을 이용하여 Support Vector Machine을 수행하며, 함수에서 사용할 수 있는 자세한 옵션은 여기를 참고한다.
ksvm(formula, data, kernel, C, kpar, prob.model, ...)
formula
: Target과 예측 변수의 관계를 표현하기 위한 함수로써 일반적으로 Target ~ 예측 변수
의 형태로 표현한다.data
: formula
에 포함하고 있는 변수들의 데이터셋(Data Frame)kernel
: Kernel 함수
"vanilladot"
: Linear Kernel \(k(\boldsymbol{x}, \boldsymbol{x}') = \boldsymbol{x}\boldsymbol{x}'\)"polydot"
: Polynomial Kernel \(k(\boldsymbol{x}, \boldsymbol{x}') = (\text{scale} * \boldsymbol{x}\boldsymbol{x}' + \text{offset})^{\text{degree}}\)"rbfdot"
: Radial Basis Kernel \(k(\boldsymbol{x}, \boldsymbol{x}') = \exp\left(-\sigma||\boldsymbol{x}-\boldsymbol{x}'||^2 \right)\)"tanhdot"
: Hyperbolic Tangent Kernel"laplacedot"
: Laplacian Kernel"splinedot"
: Spline KernelC
: 데이터를 잘못 분류하는 선을 그을 경우 지불해야 할 costkpar
: Kernel 함수에 포함된 초모수(Hyperparameter) 조합의 리스트(List)
kernel = "rbfdot"
에 대해,
sigma
: 차수prob.model
: Test Dataset
에 대한 예측 확률
의 생성 여부
TRUE
: 함수 predict()
를 이용하여 Test Dataset
에 대한 예측 확률
을 생성할 수 있다.# 옵션 kpar를 이용하여 sigma 값 입력
svm.model.rd <- ksvm(Personal.Loan ~.,
data = UB.trd,
kernel = "rbfdot",
C = 1,
kpar = list(sigma = 2),
prob.model = TRUE)
svm.model.rd
Support Vector Machine object of class "ksvm"
SV type: C-svc (classification)
parameter : cost C = 1
Gaussian Radial Basis kernel function.
Hyperparameter : sigma = 2
Number of Support Vectors : 1729
Objective Function Value : -257.0727
Training error : 0.002284
Probability model included.
# 자동으로 최적의 sigma 값 계산
svm.model.rd <- ksvm(Personal.Loan ~.,
data = UB.trd,
kernel = "rbfdot",
C = 1,
prob.model = TRUE)
svm.model.rd
Support Vector Machine object of class "ksvm"
SV type: C-svc (classification)
parameter : cost C = 1
Gaussian Radial Basis kernel function.
Hyperparameter : sigma = 0.0317835061526619
Number of Support Vectors : 284
Objective Function Value : -163.5041
Training error : 0.018846
Probability model included.
Caution!
모형 평가를 위해 Test Dataset
에 대한 예측 class/확률
이 필요하며, 함수 predict()
를 이용하여 생성한다.
# 예측 class 생성
svm.rd.pred <- predict(svm.model.rd,
newdata = UB.ted[,-1], # Test Dataset including Only 예측 변수
type = "response") # 예측 class 생성
svm.rd.pred %>%
as_tibble
# A tibble: 749 × 1
value
<fct>
1 no
2 no
3 no
4 no
5 no
6 no
7 no
8 no
9 no
10 no
# ℹ 739 more rows
CM <- caret::confusionMatrix(svm.rd.pred, UB.ted$Personal.Loan,
positive = "yes") # confusionMatrix(예측 class, 실제 class, positive="관심 class")
CM
Confusion Matrix and Statistics
Reference
Prediction no yes
no 670 22
yes 3 54
Accuracy : 0.9666
95% CI : (0.9511, 0.9783)
No Information Rate : 0.8985
P-Value [Acc > NIR] : 1.468e-12
Kappa : 0.7941
Mcnemar's Test P-Value : 0.0003182
Sensitivity : 0.7105
Specificity : 0.9955
Pos Pred Value : 0.9474
Neg Pred Value : 0.9682
Prevalence : 0.1015
Detection Rate : 0.0721
Detection Prevalence : 0.0761
Balanced Accuracy : 0.8530
'Positive' Class : yes
# 예측 확률 생성
test.svm.prob <- predict(svm.model.rd,
newdata = UB.ted[,-1], # Test Dataset including Only 예측 변수
type = "probabilities") # 예측 확률 생성
test.svm.prob %>%
as_tibble
# A tibble: 749 × 2
no yes
<dbl> <dbl>
1 0.991 0.00879
2 1.00 0.0000379
3 1.00 0.0000583
4 0.999 0.00102
5 1.00 0.0000185
6 1.00 0.000270
7 1.00 0.000165
8 0.994 0.00604
9 0.930 0.0700
10 0.999 0.000669
# ℹ 739 more rows
test.svm.prob <- test.svm.prob[,2] # "Personal.Loan = yes"에 대한 예측 확률
ac <- UB.ted$Personal.Loan # Test Dataset의 실제 class
pp <- as.numeric(test.svm.prob) # 예측 확률을 수치형으로 변환
Caution!
Package "pROC"
를 통해 출력한 ROC 곡선은 다양한 함수를 이용해서 그래프를 수정할 수 있다.
# 함수 plot.roc() 이용
plot.roc(svm.roc,
col="gray", # Line Color
print.auc = TRUE, # AUC 출력 여부
print.auc.col = "red", # AUC 글씨 색깔
print.thres = TRUE, # Cutoff Value 출력 여부
print.thres.pch = 19, # Cutoff Value를 표시하는 도형 모양
print.thres.col = "red", # Cutoff Value를 표시하는 도형의 색깔
auc.polygon = TRUE, # 곡선 아래 면적에 대한 여부
auc.polygon.col = "gray90") # 곡선 아래 면적의 색깔
# 함수 ggroc() 이용
ggroc(svm.roc) +
annotate(geom = "text", x = 0.9, y = 1.0,
label = paste("AUC = ", auc),
size = 5,
color="red") +
theme_bw()
pacman::p_load("Epi")
# install_version("etm", version = "1.1", repos = "http://cran.us.r-project.org")
ROC(pp, ac, plot = "ROC") # ROC(예측 확률, 실제 class)
pacman::p_load("ROCR")
svm.pred <- prediction(pp, ac) # prediction(예측 확률, 실제 class)
svm.perf <- performance(svm.pred, "tpr", "fpr") # performance(, "민감도", "1-특이도")
plot(svm.perf, col = "gray") # ROC Curve
perf.auc <- performance(svm.pred, "auc") # AUC
auc <- attributes(perf.auc)$y.values
legend("bottomright", legend = auc, bty = "n")
svm.perf <- performance(svm.pred, "lift", "rpp") # Lift Chart
plot(svm.perf, main = "lift curve",
colorize = T, # Coloring according to cutoff
lwd = 2)
Text and figures are licensed under Creative Commons Attribution CC BY 4.0. The figures that have been reused from other sources don't fall under this license and can be recognized by a note in their caption: "Figure from ...".