-
Notifications
You must be signed in to change notification settings - Fork 0
/
svmclassifier.R
235 lines (198 loc) · 7.57 KB
/
svmclassifier.R
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
222
223
224
225
226
227
228
229
230
231
232
233
234
235
library( 'e1071' )
library("openxlsx")
options(java.parameters = "-Xmx2048m")
setwd("C:/Users/Naseem Ashraf/Desktop/Fall 16/DM Project sets")
newproductdataset <- read.xlsx("newproductdatasetclassifiedNEW.xlsx", sheet = 1,startRow = 1, colNames = TRUE)
colnames(newproductdataset) <- c("ProductID","Description","TransactionFreq","TotalQuantity","Customers","MeanQuantityPerTransaction","MeanQuantityPerCustomer", "UnitPrice","MeanEarningPerTransaction","SalePriorityClass")
--
inx <- c(1:3877)
data1$"index" <- inx
rm(inx)
attach(data1)
#train <- as.integer(1170*0.90) #3489 ~ 90%
train = index<3489
##############################
##SVM Tuned Model 1
##############################
library(caret)
data2 <- data1
data2$"index" <- NULL
data2$"ProductID" <- NULL
data2$"Description" <- NULL
data2$"SalePriorityClass" <- as.factor(data2$"SalePriorityClass")
data2.train <- data1[data1$index<3489,]
data2.test <- data1[data1$index>=3489,]
drops <- c("SalePriorityClass","Description", "index", "ProductID")
data2.train.attributes <- data2.train[,!(names(data2.train) %in% drops)]
data2.test.attributes <- data2.test[,!(names(data2.test) %in% drops)]
# str(data2.train.attributes)
# str(data2.test.attributes)
data2class.train <- as.factor(data2.train$SalePriorityClass)
data2class.test <- as.factor(data2.test$SalePriorityClass)
svm_model1 <- svm(data2.train.attributes, data2class.train)
svm.original.pred1 <- predict(svm_model1, data2.test.attributes)
tab <- table(pred = svm.original.pred1, true = data2class.test)
tab
# true
# pred 1 2 3
# 1 5 0 1
# 2 29 52 1
# 3 0 2 299
classification_error_original1 <- 1- sum(svm.original.pred1 == data2class.test)/length(svm.original.pred1)
classification_error_original1
# 0.0848329
svm_tune1 <- tune(svm, train.x = data2.train.attributes, train.y = data2class.train,
kernel="radial", ranges = list(cost=10^-1:2), gamma=c(.5,1,2))
svm_tune1$best.model
# Call:
# best.tune(method = svm, train.x = data2.train.attributes, train.y = data2class.train, ranges = list(cost = 10^-1:2),
# kernel = "radial", gamma = c(0.5, 1, 2))
#
#
# Parameters:
# SVM-Type: C-classification
# SVM-Kernel: radial
# cost: 1.1
# gamma: 0.5 1 2
#
# Number of Support Vectors: 659
svm_model1_best <- svm(data2.train.attributes, data2class.train, kernel="radial", cost=1.1, gamma=c(0.5,1,2))
svm.pred1 <- predict(svm_model1_best, data2.test.attributes)
tab <- table(pred = svm.pred1, true = data2class.test)
tab
# true
# pred 1 2 3
# 1 9 1 1
# 2 25 52 1
# 3 0 1 299
classification_error1 <- 1- sum(svm.pred1 == data2class.test)/length(svm.pred1)
classification_error1
# 0.07455013
##############################
##SVM Tuned Model 2
##############################
library(caret)
## SVM Model 2
data2 <- data1
data2$"index" <- NULL
data2$"ProductID" <- NULL
data2$"Description" <- NULL
data2$"SalePriorityClass" <- as.factor(data2$"SalePriorityClass")
data2.train <- data1[data1$index<3489,]
data2.test <- data1[data1$index>=3489,]
# cbind(TransactionFreq, Customers, MeanQuantityPerTransaction,
# MeanEarningPerTransaction)
drops <- c("ProductID","SalePriorityClass","Description", "index", "TotalQuantity", "MeanQuantityPerCustomer",
"UnitPrice")
data2.train.attributes <- data2.train[,!(names(data2.train) %in% drops)]
data2.test.attributes <- data2.test[,!(names(data2.test) %in% drops)]
names(data2.train.attributes)
str(data2.train.attributes)
str(data2.test.attributes)
data2class.train <- as.factor(data2.train$SalePriorityClass)
data2class.test <- as.factor(data2.test$SalePriorityClass)
svm_model2 <- svm(data2.train.attributes, data2class.train)
svm.original.pred2 <- predict(svm_model2, data2.test.attributes)
tab <- table(pred = svm.original.pred2, true = data2class.test)
tab
# true
# pred 1 2 3
# 1 7 1 0
# 2 27 52 1
# 3 0 1 300
classification_error_original2 <- 1- sum(svm.original.pred2 == data2class.test)/length(svm.original.pred2)
classification_error_original2
# 0.07712082
svm_tune2 <- tune(svm, train.x = data2.train.attributes, train.y = data2class.train,
kernel="radial", ranges = list(cost=10^-1:2), gamma=c(.5,1,2))
svm_tune2$best.model
# Call:
# best.tune(method = svm, train.x = data2.train.attributes, train.y = data2class.train, ranges = list(cost = 10^-1:2),
# kernel = "radial", gamma = c(0.5, 1, 2))
#
#
# Parameters:
# SVM-Type: C-classification
# SVM-Kernel: radial
# cost: 1.1
# gamma: 0.5 1 2
#
# Number of Support Vectors: 718
svm_model2_best <- svm(data2.train.attributes, data2class.train, kernel="radial", cost=1.1, gamma=c(0.5,1,2))
svm.pred2 <- predict(svm_model2_best, data2.test.attributes)
tab <- table(pred = svm.pred2, true = data2class.test)
tab
# true
# pred 1 2 3
# 1 6 1 0
# 2 28 53 1
# 3 0 0 300
classification_error2 <- 1- sum(svm.pred2 == data2class.test)/length(svm.pred2)
classification_error2
# 0.07712082
##############################
##SVM Tuned Model 3
##############################
library(caret)
## SVM Model 3
data2 <- data1
data2$"index" <- NULL
data2$"ProductID" <- NULL
data2$"Description" <- NULL
data2$"SalePriorityClass" <- as.factor(data2$"SalePriorityClass")
data2.train <- data1[data1$index<3489,]
data2.test <- data1[data1$index>=3489,]
# cbind(TransactionFreq, Customers, MeanEarningPerTransaction)
drops <- c("SalePriorityClass","Description", "index", "TotalQuantity", "MeanQuantityPerCustomer",
"UnitPrice", "MeanQuantityPerTransaction", "ProductID")
data2.train.attributes <- data2.train[,!(names(data2.train) %in% drops)]
data2.test.attributes <- data2.test[,!(names(data2.test) %in% drops)]
names(data2.train.attributes)
str(data2.train.attributes)
str(data2.test.attributes)
data2class.train <- as.factor(data2.train$SalePriorityClass)
data2class.test <- as.factor(data2.test$SalePriorityClass)
svm_model3 <- svm(data2.train.attributes, data2class.train)
svm.original.pred3 <- predict(svm_model3, data2.test.attributes)
tab <- table(pred = svm.original.pred3, true = data2class.test)
tab
# true
# pred 1 2 3
# 1 7 1 0
# 2 27 52 1
# 3 0 1 300
classification_error_original3 <- 1- sum(svm.original.pred3 == data2class.test)/length(svm.original.pred3)
classification_error_original3
# 0.07712082
svm_tune3 <- tune(svm, train.x = data2.train.attributes, train.y = data2class.train,
kernel="radial", ranges = list(cost=10^-1:2), gamma=c(.5,1,2))
svm_tune3$best.model
# Call:
# best.tune(method = svm, train.x = data2.train.attributes, train.y = data2class.train, ranges = list(cost = 10^-1:2),
# kernel = "radial", gamma = c(0.5, 1, 2))
#
#
# Parameters:
# SVM-Type: C-classification
# SVM-Kernel: radial
# cost: 1.1
# gamma: 0.5 1 2
svm_model3_best <- svm(data2.train.attributes, data2class.train, kernel="radial", cost=1.1, gamma=c(0.5,1,2))
svm.pred3 <- predict(svm_model3_best, data2.test.attributes)
tab <- table(pred = svm.pred3, true = data2class.test)
tab
# true
# pred 1 2 3
# 1 6 1 0
# 2 28 53 1
# 3 0 0 300
classification_error3 <- 1- sum(svm.pred3 == data2class.test)/length(svm.pred3)
classification_error3
# 0.07712082
#SVM Model performance
errors <- c(classification_error1*100, classification_error2*100, classification_error3*100)
barplot(errors, ylim = c(6,11), beside=TRUE, xpd = FALSE, ylab = "Error %", names.arg=c(1,2,3),
xlab = "SVM Model Number")
text(0.70, 8, format(classification_error1*100))
text(1.85, 8, format(classification_error2*100))
text(3.00, 8, format(classification_error3*100))