-
Notifications
You must be signed in to change notification settings - Fork 1
/
SVM-UI2year.Rmd
221 lines (155 loc) · 7.39 KB
/
SVM-UI2year.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
---
title: "SVM for urine loss 2-year"
author: "Hajar Hasannejadasl"
date: "2022-09-02"
output: html_document
---
```{r setup, include=FALSE}
library(classifierplots)
library(caret)
knitr::opts_chunk$set(echo = TRUE)
setwd("C:/Users/hajar.hasannejadasl/Documents/prospect2022/")
scriptsLocations = paste(getwd(), "/scripts/", sep = "")
source("scripts/importAllHandmadeFunctions.R")
importAllHandmadeFunctions(scriptsLocations)
#Build 1-year prediction model for urine loss with SVM
#load data
allData2years <- readRDS("data/allData2years")
```
```{r}
training= data.frame(allData2years[1])
training=training[,1:38]
test= data.frame(allData2years[2])
test=test[,1:38]
```
Before we start, we’ll first remove the variables related to hot flashes and sensitive breasts as these are not to be included upon request by collaborators.
```{r}
training = training[,-grep("epic26_22_opvliegers1", names(training))]
training = training[,-grep("epic26_23_gevoeligeborsten1", names(training))]
test = test[,-grep("epic26_22_opvliegers1", names(test))]
test = test[,-grep("epic26_23_gevoeligeborsten1", names(test))]
```
First we recreate the training and test set with only the input variables and the outcome we want, in this case *“epic26_1_urineverlies2”. We’ll also remove the outcome data that we’re not interested in right now from the training and test set.
```{r}
outcomeTraining = allData2years[[1]]$epic26_1_urineverlies3
outcomeTest = allData2years[[2]]$epic26_1_urineverlies3
```
Now we’ll make our binary data set
```{r}
#convert to binary
BinaryoutcomeTest<-ifelse(outcomeTest<5, "1","0")
BinaryoutcomeTraining<-ifelse(outcomeTraining<5, "1","0")
#convert to factor
BinaryoutcomeTraining=as.numeric(BinaryoutcomeTraining)
BinaryoutcomeTest=as.factor(BinaryoutcomeTest)
```
Prediction after Upsampling of the smaller dataset
Upsampling on class 2 (patients with problems) on train data
```{r}
#Upsampling
library(smotefamily)
trainingSet2 <- data.frame(cbind(training,BinaryoutcomeTraining))
trainingSet2$BinaryoutcomeTraining <- as.factor(trainingSet2$BinaryoutcomeTraining)
smote <- smotefamily::SMOTE(trainingSet2[,-which(colnames(trainingSet2) == "BinaryoutcomeTraining")], trainingSet2$BinaryoutcomeTraining, K=5, dup_size = 1)
datasmote=smote$data
table(datasmote$class)
SMOTE2BinaryOutcomeTraining = (datasmote[,c(37)])
SMOTE2OutcomeTraining = (datasmote[,c(37)])
SMOTE2TrainingSet = datasmote[,-c(37)]
SMOTE2BinaryOutcomeTraining<-as.factor(SMOTE2BinaryOutcomeTraining)
SMOTE2OutcomeTraining<-as.factor(SMOTE2BinaryOutcomeTraining)
```
The SMOTE function has generated 214 syntheic data. The dataset is now almost fully balanced.
#Recursive feature elimination
```{r}
levels(SMOTE2BinaryOutcomeTraining) <- c("firstclass", "secondclass")
# svmProfile_linear <- rfe(SMOTE2TrainingSet, SMOTE2BinaryOutcomeTraining,
# sizes = c(2:36),
# rfeControl = rfeControl(functions = caretFuncs,
# number = 5),
#
# method = "svmLinear", metric = "Accuracy")
#
#
# saveRDS(svmProfile_linear, "C:/Users/hajar.hasannejadasl/Documents/prospect2022/models/SVMProfile-UI-2year.rds")
svmProfile_linear = readRDS("C:/Users/hajar.hasannejadasl/Documents/prospect2022/models/SVMProfile-UI-2year.rds")
```
We’ll now run SVM models based on the variable importance retrieved by RFE.
```{r}
varImportance = calculateVariableImportance(svmProfile_linear)
```
```{r, analysis, message=FALSE, results='markup'}
allSVMModels = reduceVariablesLogit(SMOTE2TrainingSet, SMOTE2BinaryOutcomeTraining, varImportance)
```
```{r}
{plot(rev(allSVMModels[[2]]$numberOfVariables), rev(allSVMModels[[2]]$AUC), ylim = c(0, 1), pch = 16)
points(rev(allSVMModels[[2]]$numberOfVariables), rev(allSVMModels[[2]]$sensitivity), col = "red", pch = 16)
points(rev(allSVMModels[[2]]$numberOfVariables), rev(allSVMModels[[2]]$specificity), col = "blue", pch = 16)
points(rev(allSVMModels[[2]]$numberOfVariables), rev(allSVMModels[[2]]$overallAccuracy), col = "green", pch = 16)
abline(v=12, col = "red")
legend("bottomright", legend = c("AUC", "sensitivity", "specificity", "overall accuracy"), col = c("black", "red", "blue", "green"), pch = 16)}
```
```{r}
#Checking the performance metrics
allSVMModels[2]
```
We chose the model with 9 variables.
```{r}
#ChosenModel
chosenmodelvariables<-svmProfile_linear$optVariables[1:9]
chosenmodelvariables
```
Now we want know how this model performs on our training set first.
```{r}
newdataset<-SMOTE2TrainingSet[c(chosenmodelvariables)]
#train with chosen variables
set.seed(350)
# Fit the model
combined<-cbind(newdataset,SMOTE2BinaryOutcomeTraining)
# model_svmRFE<- train(SMOTE2BinaryOutcomeTraining ~ ., data = combined,
# method = "svmLinear",
# preProcess = c("center", "scale"),
# trControl = trainControl(method = "boot", number = 25, classProbs = TRUE, summaryFunction = twoClassSummary, savePredictions = TRUE))
# saveRDS(model_svmRFE, "C:/Users/hajar.hasannejadasl/Documents/prospect2022/models/model_svmRFE-2year.rds")
model_svmRFE = readRDS("C:/Users/hajar.hasannejadasl/Documents/prospect2022/models/model_svmRFE-2year.rds")
#Confusion matrix for train data
predict_svmRFE_train<- predict(model_svmRFE, newdata = combined,na.action = na.pass)
confusionMatrix(table(predict_svmRFE_train, SMOTE2BinaryOutcomeTraining))
```
```{r}
#ROC plot
train_prob = predict(model_svmRFE, SMOTE2TrainingSet, type = "prob")
roc_svm_one <- roc(SMOTE2BinaryOutcomeTraining, as.vector(train_prob[,1]))
plot(roc_svm_one, col = "black",lty = 1,cex.lab=1.5, cex.axis=1.5, lwd = 3, print.auc=TRUE)
#calibration plot
calibration_plot(SMOTE2OutcomeTraining, as.vector(train_prob[,1]))
```
```{r}
# Linear SVM Performance Evaluation in test set
testcombine<- cbind(test,BinaryoutcomeTest)
testcombine$BinaryoutcomeTest<-as.factor(testcombine$BinaryoutcomeTest)
levels(testcombine$BinaryoutcomeTest) <- c("firstclass", "secondclass")
predict_svmRFE_test <- predict(model_svmRFE, newdata = testcombine,na.action = na.pass)
confusionMatrix(table(predict_svmRFE_test, testcombine$BinaryoutcomeTest))
```
```{r}
test_probSVM = predict(model_svmRFE, testcombine, type = "prob")
roc_svm_one <- roc(BinaryoutcomeTest, as.vector(test_probSVM[,1]))
plot(roc_svm_one, col = "black",lty = 1,cex.lab=1.5, cex.axis=1.5, lwd = 3, print.auc=TRUE)
```
Important Variables
```{r}
imp <- (varImp(svmProfile_linear))
imp$Overall<-round(imp$Overall*100, digit=2)
varimp_data <- data.frame(feature = row.names(imp),
importance = imp$Overall)
varimp_data2<-data.frame(varimp_data[1:9,])
#rename rows to English
varimp_data2$feature2 <- c("Treatments","sCT","Leaking urine", "Urinary control", "Urinary function", "Urinate frequency","Urine loss", "Gleason group", "sCN")
ggplot(data = varimp_data2,
aes(x = reorder(feature2, importance), y = importance, fill = "#7D3C98")) +
coord_flip() +
geom_bar(stat="identity", fill="#7D3C98") + labs(x = "Features", y = "Variable Importance") +
geom_text(aes(label = round(importance,2)), vjust=0, hjust=0, color="black", size=3) +
theme_bw() + theme(legend.position = "none")
```