-
Notifications
You must be signed in to change notification settings - Fork 0
/
ThreeBank.R
208 lines (187 loc) · 6.42 KB
/
ThreeBank.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
#################
###############
## Ucitivamo dataSet
data<-read.csv("dataSets/bank.csv",stringsAsFactors = F)
rm(chr_vars)
rm(num_vars)
### Kreiranje izlazne varijable
any(is.na(data$Exited))
data$Stayed<-ifelse(data$Exited==0,"yes","no")
data$Stayed<-as.factor(data$Stayed)
data$Exited<-NULL
#### Napraviti podskup koji ne sadrzi klijente starije od 87 godina
any(is.na(data$Age))
## Sredjivanje varijable Age
sum(data$Age=="" | data$Age=="-" | data$Age=="-",na.rm = T)
data$Age[data$Age=="" | data$Age=="-" | data$Age=="-"]<-NA
data$Age<-as.numeric(data$Age)
shapiro.test(data$Age)
data$Age[is.na(data$Age)]<-median(data$Age,na.rm = T)
dataSub<-subset(data,Age<87)
data<-dataSub
rm(dataSub)
### Procena atributa za model naivnog bayesa ###
unique(data$Surname)
## Surname -iskljuciti jer ima previse razlicith vrednosti
data$Surname<-NULL
chr_vars<-c("Geography","Gender","Card.Type")
apply(data[,chr_vars],2,function(x) sum(x=="" | x=="-" | x==" ",na.rm = T))
data$Geography[data$Geography=="" | data$Geography=="-" | data$Geography==" "]<-NA
### Sredjivanje numerickih varijabli ###
num_vars<-c(1,4:11,13)
apply(data[,num_vars],2,function(x) sum(is.na(x)))
#### Pretvaranje u numeric tip sve
data[,num_vars]<-apply(data[,num_vars],2,function(x) as.numeric(x))
######## Geography ######
any(is.na(data$Geography))
unique(data$Geography)
table(data$Geography)
data$Geography[is.na(data$Geography)]<-"France"
data$Geography<-as.factor(data$Geography)
########### Gender #####
unique(data$Gender)
data$Gender<-as.factor(data$Gender)
########### Card.Type #####
unique(data$Card.Type)
data$Card.Type<-factor(data$Card.Type,levels = c("SILVER","GOLD","PLATINUM","DIAMOND"))
data$HasCrCard<-factor(data$HasCrCard,levels = 0:1,labels = c("no","yes"))
data$IsActiveMember<-factor(data$IsActiveMember,levels = 0:1,labels = c("no","yes"))
###############
which(complete.cases(data)==F)
###############
library(ggplot2)
###############
apply(data[,num_vars],2,shapiro.test)
##### Credit Score ###
ggplot(data,aes(x=CreditScore,fill=Stayed))+
geom_density(alpha=0.55)+
theme_classic()
# Na prvi pogled izgleda da ne postoji statisticka znacajna razlika izmedju ovih oblati
wilcox.test(data$CreditScore[data$Stayed=="yes"],data$CreditScore[data$Stayed=="no"])
#### ne postoji statisticka znacajna razlika pa je odbacujemo
data$CreditScore<-NULL
#### Age ####
ggplot(data,aes(x=Age,fill=Stayed))+
geom_density(alpha=0.55)+
theme_classic()
#### postoji statisticka znacajna razlika izmedju ove dve oblasti pa uzimamo ovu promenljivu
#### Tanure #####
ggplot(data,aes(x=Tenure,fill=Stayed))+
geom_density(alpha=0.55)+
theme_classic()
# Na prvi pogled izgleda da ne postoji statisticka znacajna razlika izmedju ovih oblati
wilcox.test(data$Tenure[data$Stayed=="yes"],data$Tenure[data$Stayed=="no"])
### ne postoji statisticki znacajna razlika izmedju ove dve oblaste pa je odbacujemo
data$Tenure<-NULL
#### Balance #####
ggplot(data,aes(x=Balance,fill=Stayed))+
geom_density(alpha=0.55)+
theme_classic()
### postoji statisticki znacajna razlika, prihvatamo prom
#### Num of products ####
ggplot(data,aes(x=NumOfProducts,fill=Stayed))+
geom_density(alpha=0.55)+
theme_classic()
### postoji statisticki znacajna razlika, prihvatamo
#### Estimated Salary
ggplot(data,aes(x=EstimatedSalary,fill=Stayed))+
geom_density(alpha=0.55)+
theme_classic()
# Na prvi pogled izgleda da ne postoji statisticka znacajna razlika izmedju ovih oblati
wilcox.test(data$EstimatedSalary[data$Stayed=="yes"],data$EstimatedSalary[data$Stayed=="no"])
### ne postoji statisticki znacajna razlika, ovu promenljivu necemo uzimati
data$EstimatedSalary<-NULL
#### Satisfactional score ###
ggplot(data,aes(x=Satisfaction.Score,fill=Stayed))+
geom_density(alpha=0.55)+
theme_classic()
wilcox.test(data$Satisfaction.Score[data$Stayed=="yes"],data$Satisfaction.Score[data$Stayed=="no"])
### ne postoji statisticki znacajna razlika izmedju ovih promenljivi, pa je odbacujemo
data$Satisfaction.Score<-NULL
### Point Earned ###
ggplot(data,aes(x=Point.Earned,fill=Stayed))+
geom_density(alpha=0.55)+
theme_classic()
### na prvi pogled ne postoji statisticki znacajna razlika
wilcox.test(data$Point.Earned[data$Stayed=="yes"],data$Point.Earned[data$Stayed=="no"])
data$Point.Earned<-NULL
############################
### Geography ###
ggplot(data,aes(x=Geography,fill=Stayed))+
geom_bar(position = "fill")+
theme_classic()
### Gender ####
ggplot(data,aes(x=Gender,fill=Stayed))+
geom_bar(position = "fill")+
theme_classic()
### prihvatamo ovu jer nisu iste proporcije
### HasCrCard ####
ggplot(data,aes(x=HasCrCard,fill=Stayed))+
geom_bar(position = "fill")+
theme_classic()
### ista je proporcija, pa cemo ovu varijablu odbaciti
data$HasCrCard<-NULL
#### IsActiveMember ####
ggplot(data,aes(x=IsActiveMember,fill=Stayed))+
geom_bar(position = "fill")+
theme_classic()
### prihvatamo jer ne postoje iste proporcije ###
#### Card Type ####
ggplot(data,aes(x=Card.Type,fill=Stayed))+
geom_bar(position = "fill")+
theme_classic()
data$Card.Type<-NULL
######## ALGORITAM ############
##### Podela na traing i test #####
library(caret)
set.seed(4623)
train_ind<-createDataPartition(data$Stayed,p=0.8,list = F)
train<-data[train_ind,]
test<-data[-train_ind,]
###########
library(rpart)
library(rpart.plot)
set.seed(4623)
tree1<-rpart(Stayed~.,data = train)
rpart.plot(tree1)
tree1_pred<-predict(tree1,newdata = test,type = "class")
cm1<-table(actual=test$Stayed,predicted=tree1_pred)
cm1
#######################
compute_eval_metrics<-function(cmatrix){
TP<-cmatrix[2,2]
TN<-cmatrix[1,1]
FP<-cmatrix[1,2]
FN<-cmatrix[2,1]
a<-(TP+TN)/sum(cmatrix)
p<-TP/(TP+FP)
r<-TP/(TP+FN)
f1<-2*p*r/(p+r)
c(accuracy=a,precision=p,recall=r,F1=f1)
}
##########################
eval1<-compute_eval_metrics(cm1)
eval1
###########################
#### Kros validacija #####
library(e1071)
tr_Control<-trainControl(method = "cv",number = 10)
tune_Grid<-expand.grid(.cp=seq(0.001,0.05,0.005))
set.seed(4623)
tree_cv=train(x=train[,-7],
y=train$Stayed,
method="rpart",
trControl=tr_Control,
tuneGrid=tune_Grid,
minsplit=10)
plot(tree_cv)
best_cp<-tree_cv$bestTune$cp
##########################
set.seed(4623)
tree2<-rpart(Stayed~.,data = train,control = rpart.control(minsplit = 10,cp=best_cp))
rpart.plot(tree2)
tree2_pred<-predict(tree2,newdata = test,type = "class")
cm2<-table(actual=test$Stayed,predicted=tree2_pred)
cm2
eval2<-compute_eval_metrics(cm2)
data.frame(rbind(eval1,eval2))