-
Notifications
You must be signed in to change notification settings - Fork 3
/
Model_Trees.R
354 lines (271 loc) · 11.7 KB
/
Model_Trees.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
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
#### LOADING FUNCTIONS & DATA ####
rm(list = ls())
library(forcats)
library(FNN) # used
library(varhandle) # used
library(rlang)
library(Rtsne)
library(ROCR)
library(VIM)
library(PresenceAbsence) # used
library(stringr)
library(caret) # used
library(bit64) # used
library(caTools) #used
library(fmsb)
library(RSKC) #used
library(sparcl) #used alternatively to RSKC
library(factoextra)
library(cluster)
library(NbClust)
library(Fdplyr::selector) # used
library(data.table) # used
library(compare)
library(devtools) # used
library(mice) # used
library(Boruta) # used
library(smotefamily) # used
library(readxl)
library(tidyverse) # used
environment(arrange)
# Function that identifies if the variable can be used as factor
is.whole = function(x){
ifelse(all(is.character(x)),FALSE,all(ifelse(as.integer(x)%%1==0,TRUE,FALSE),na.rm = TRUE))
}
can.factor = function(x){
ifelse(is.character(x)|is.logical(x)|(length(unique(x))<15&is.whole(x)),TRUE,FALSE)
}
# VIF: Calculates the VIF
vif = function(df){
diag(solve(cor(df,use = 'complete.obs')))
}
num = function(col){
as.numeric(gsub(",", "", col))
}
# Extraction tool
extr = function(scenario){
df=data_scenario[data_scenario$scenario==scenario,] # filtering the main Scenario DF on dplyr::selected scenario
df=df[, colSums(df != "") != 0] # Eliminating empty columns
df$scenario=NULL # Just droping this which is already redundant
df=apply(df,2,num)%>%as_data_frame() # making all characters / factors into numeric
colnames(df)[-1] = paste(colnames(df)[-1], deparse(scenario), sep = "_") # Adding suffix depending on the scenario
data2 = left_join(data3,df,by = 'index')
assign("data3", data2, envir = .GlobalEnv)
}
multi.fun = function(x) {
c(mean = mean(x),median=median(x), sd=sd(x), mad=mad(x),
IQR = IQR(x),
var.coef=(sd(x)/mean(x)),
var.coef2=(IQR(x)/median(x)),
max = max(x), min= min(x),
skewness= mean((x- mean(x))^3)/(sd(x)^3)
)
}
# Mode Function
Mode = function(x, na.rm = FALSE) {
if(na.rm){
x = x[!is.na(x)]
}
ux = unique(x)
return(ux[which.max(tabulate(match(x, ux)))])
}
Mode2 = function(x) {
ux = unique(x)
if(!anyDuplicated(x)){
NA_character_ } else {
tbl = tabulate(match(x, ux))
toString(ux[tbl==max(tbl)])
}
}
median_mode = function(x){
ifelse(is.numeric(x),median(x),Mode2(x))
}
# Load data properly
setwd("C:/Users/orteg/Documents/GitHub/datathon/Credit")
df = read_excel("Credit_DataSet.xlsx", sheet = 2,
na=c(""," ",".","NA","#N/A",'#N/A N/A')) # so that it interprets blank spaces as NA's
#### DATA CLEANING ####
glimpse(df)
df$ID = NULL
df$COUNTRY_NAME = NULL
# Currency
dfaux = df[is.na(df$CURRENCY),]
# Manual Lookup [Daniel]
dfaux[dfaux$COUNTRY=="AT",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="BE",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="BM",]$CURRENCY = "BMD"
dfaux[dfaux$COUNTRY=="CA",]$CURRENCY = "CAD"
dfaux[dfaux$COUNTRY=="CH",]$CURRENCY = "CHF"
dfaux[dfaux$COUNTRY=="CN",]$CURRENCY = "CNY"
dfaux[dfaux$COUNTRY=="DE",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="DK",]$CURRENCY = "DKK"
dfaux[dfaux$COUNTRY=="ES",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="FI",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="FR",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="GB",]$CURRENCY = "GBP"
dfaux[dfaux$COUNTRY=="GR",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="IE",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="IL",]$CURRENCY = "ILS"
dfaux[dfaux$COUNTRY=="IN",]$CURRENCY = "INR"
dfaux[dfaux$COUNTRY=="IT",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="JE",]$CURRENCY = "GBP"
dfaux[dfaux$COUNTRY=="JP",]$CURRENCY = "JPY"
dfaux[dfaux$COUNTRY=="KY",]$CURRENCY = "KYD"
dfaux[dfaux$COUNTRY=="LU",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="MC",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="NL",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="NO",]$CURRENCY = "NOK"
dfaux[dfaux$COUNTRY=="PR",]$CURRENCY = "USD"
dfaux[dfaux$COUNTRY=="PT",]$CURRENCY = "EUR"
dfaux[dfaux$COUNTRY=="SE",]$CURRENCY = "SEK"
dfaux[dfaux$COUNTRY=="TH",]$CURRENCY = "THB"
dfaux[dfaux$COUNTRY=="US",]$CURRENCY = "USD"
dfaux[dfaux$COUNTRY=="ZA",]$CURRENCY = "ZAR"
# Add again on main df
df = rbind(dfaux, df[!is.na(df$CURRENCY),])
df$CURRENCY = as.factor(df$CURRENCY)
#Change strange names to euros
df$CURRENCY[df$CURRENCY %in% c('EDM','EES','EIL','EFR','EIP','EFM','EDG','EPE','EAS','EBF','EGD','EUR')] = 'EUR'
df.stand = df
df.stand$MARKETCAP = df.stand$MARKETCAP/1000000
df.stand[,c(22,24:30)] = df.stand[,c(22,24:30)]/df.stand$TOTAL_ASSETS
df.stand$TOTAL_ASSETS = NULL
df = df.stand
rm(df.stand)
# Missing Values
df$PCT_WOMEN_EMPLOYEES = NULL
df$PCT_WOMEN_MGT = NULL
df$ETHICS_POLICY[is.na(df$ETHICS_POLICY)] = 'N'
df$WHISTLE_BLOWER_POLICY[is.na(df$WHISTLE_BLOWER_POLICY)] = 'N'
df$BRIBERY_POLICY[is.na(df$BRIBERY_POLICY)] = 'N'
# Missing Plot on DF [Warning: Data Leaking]
missingness = aggr(df, col=c('navyblue','yellow'),
numbers=TRUE, sortVars=TRUE,
labels=names(df_tr), cex.axis=.7,
gap=3, ylab=c("Missing data","Pattern"))
# Missing Values
df$EMPL_GROWTH[is.na(df$EMPL_GROWTH)] = median(df$EMPL_GROWTH,na.rm = TRUE)
df$PCT_CHG_1_YEAR[is.na(df$PCT_CHG_1_YEAR)] = median(df$PCT_CHG_1_YEAR,na.rm = TRUE)
df$VOLATILITY_180D[is.na(df$VOLATILITY_180D)] = median(df$VOLATILITY_180D,na.rm = TRUE)
df$PCT_CHG_6_M[is.na(df$PCT_CHG_6_M)] = median(df$PCT_CHG_6_M,na.rm = TRUE)
#### DATA WRANGLING ####
## DATA TRANSFORMATION: Making seemingly numeric variables as such, others can be factors
vector_attr = attributes(which(apply(apply(df,2,check.numeric),2,all)))[['names']] # Identify numeric-able
df = df %>% mutate_if(names(df) %in% vector_attr,as.numeric) # Make it numeric if numeric-able
df = df %>% mutate_if(can.factor,as.factor) # Make factor if factor or character
glimpse(df)
# index
set.seed(123); index = sample(1:nrow(df), size = floor(nrow(df)*0.70))
# training sets
df_tr = df[index,]
df_ts = df[-index,]
#### EDA ####
# Employment Growth Outliers
multi.fun(df$EMPL_GROWTH[complete.cases(df$EMPL_GROWTH)])
boxplot2(df$EMPL_GROWTH)
# It seems that Market Cap (Equity) + Liabilities ~ Assets (Accounting Equation)
# So it seems that there's multicollinearity, maybe PCA can help
df_tr %>% dplyr::select_if(is.numeric) %>% cor(use='complete.obs') %>% corrplot::corrplot()
# Checking if there are redundant variables with only one factor
sapply(df_tr[names(dplyr::select_if(df_tr[,-c(1:2)], is.factor))],levels)
glimpse(df_tr)
# t-SNE: Need to take only numeric data
tsne_df = df_tr %>% select_if(is.numeric) %>% scale() %>% as_data_frame()
set.seed(123); tsne = Rtsne(tsne_df,dims = 2,
perplexity = 30, verbose = TRUE,
check_duplicates = F, max_iter = 5000)
set.seed(123); tsne_3D = Rtsne(tsne_df,dims = 3,
perplexity = 30, verbose = TRUE,
check_duplicates = F, max_iter = 5000)
tsne_3D_tr = as_data_frame(tsne_3D$Y)
tsne_2D_tr = as_data_frame(tsne$Y)
df_tr %>% dplyr::select_if(is.numeric) %>% bind_cols(tsne_3D_tr) %>%
cor(use='complete.obs') %>% corrplot::corrplot()
df_tr %>% dplyr::select_if(is.numeric) %>% bind_cols(tsne_2D_tr) %>%
cor(use='complete.obs') %>% corrplot::corrplot()
# Plots t-SNE 2D
ggplot(bind_cols(PROB=df_tr$DEFAULT_PROB,tsne_2D_tr), aes(x=V1, y=V2)) +
geom_point(size=1, aes(col = PROB)) +
xlab("") + ylab("") +
ggtitle("Graph 4: t-SNE") +
theme(plot.title = element_text(hjust = 0.5))
# 3D Plot
plot_ly(data.table(tsne_3D_tr), x = ~V1, y = ~V2, z = ~V3,
color = ~tsne_df$DEFAULT_PROB, colors = c('#FFE1A1', '#683531')) %>%
add_markers() %>% layout(scene = list(xaxis = list(title = 'V1'),
yaxis = list(title = 'V2'),
zaxis = list(title = 'V3')))
rm(tsne_df)
#### POST-EDA PREPROCESSING ####
df_tr = df_tr %>% dplyr::select(DEFAULT_PROB, everything())
scl_tr_center = attributes(scale(df_tr[,-1] %>% select_if(is.numeric)))[['scaled:center']]
scl_tr_scale = attributes(scale(df_tr[,-1] %>% select_if(is.numeric)))[['scaled:scale']]
df_tr = df_tr[,-1] %>% select_if(is.numeric) %>% scale() %>%
cbind(DEFAULT_PROB=df_tr$DEFAULT_PROB, select_if(df_tr,is.factor)) %>%
as_data_frame() # Create new dataframe for scaling
#### CLUSTER ANALYSIS ####
# Decide the number of clusters
df_tr_num = df_tr %>% select_if(is.numeric) %>% as.matrix()
# set.seed(123); clest = Clest(df_tr_num, maxK = 3, alpha = 0.1, B = 15, B0 = 5, nstart = 1000)
# Calculate the Regularization Parameter
set.seed(123); kperm_4 = KMeansSparseCluster.permute(df_tr_num, K=4, nperms = 5)
# Cluster
set.seed(123); rskm_4 = RSKC(d = df_tr_num, ncl = 4, alpha = 0.1, L1 = kperm_4$bestw)
# Visualization of Clusters
fviz_cluster(list(data = df_tr_num, cluster = rskm_4$labels),
stand = FALSE, geom = "point",frame.type = "norm")
fviz_cluster(list(data = tsne_2D_tr, cluster = rskm_4$labels),
stand = FALSE, geom = "point",frame.type = "norm")
# Important Variables for Clustering
cluster_vars = data.frame('Variables' = unlist(attributes(rskm_4$weights)),'Weights' = rskm_4$weights) %>%
arrange(desc(Weights)) %>% filter(Weights > 0.01)
# Interprete
cl_interprete = df_tr %>% dplyr::select(as.character(cluster_vars$Variables))
cl_interprete_all = df_tr
cl_interprete_all$cluster = as.factor(rskm_4$labels)
cl_interprete$cluster = as.factor(rskm_4$labels)
cl_interprete = dplyr::select(cl_interprete, cluster, everything())
profile = apply(cl_interprete[,-1], 2, function(x) tapply(x, cl_interprete$cluster, median_mode))
table(rskm_4$labels)
# Add cluster into df
df_tr$labels = as.factor(rskm_4$labels)
#### OUTLIER DETECTION ####
df_tr_1 = df_tr %>% filter(labels == 1)
df_tr_2 = df_tr %>% filter(labels == 2)
df_tr_3 = df_tr %>% filter(labels == 3)
df_tr_4 = df_tr %>% filter(labels == 4)
isotree_1 = IsolationForest::IsolationTrees(x = df_tr_1 %>% dplyr::select(-labels))
isotree_2 = IsolationForest::IsolationTrees(x = df_tr_2 %>% dplyr::select(-labels))
isotree_3 = IsolationForest::IsolationTrees(x = df_tr_3 %>% dplyr::select(-labels))
isotree_4 = IsolationForest::IsolationTrees(x = df_tr_4 %>% dplyr::select(-labels))
df_tr_1$isof = IsolationForest::AnomalyScore(df_tr_1 %>% dplyr::select(-labels),isotree_1)[['outF']]
df_tr_2$isof = IsolationForest::AnomalyScore(df_tr_2 %>% dplyr::select(-labels),isotree_2)[['outF']]
df_tr_3$isof = IsolationForest::AnomalyScore(df_tr_3 %>% dplyr::select(-labels),isotree_3)[['outF']]
df_tr_4$isof = IsolationForest::AnomalyScore(df_tr_4 %>% dplyr::select(-labels),isotree_4)[['outF']]
df_tr2 = bind_rows(df_tr_1,df_tr_2,df_tr_3,df_tr_4)
df_tr2$labels = NULL
#### FEATURE SELECTION ####
set.seed(123); bor = Boruta(DEFAULT_PROB~., data = df_tr2, doTrace = 2)
plot(bor,cex.axis=.6, las=2, xlab="", main="Variable Importance")
featlist = attStats(bor) %>% rownames_to_column(var = 'Features') %>%
dplyr::select(Features, medianImp) %>% mutate(rank = rank(-medianImp)) %>% arrange(rank)
df_tr_bor = df_tr2 %>% dplyr::select(DEFAULT_PROB, featlist$Features[c(1:10)])
vif(df_tr_bor)
df_tr$isof
#### MODEL BUILDING ####
train_control = trainControl(method="cv",number=10,savePredictions=TRUE)
lm = train(DEFAULT_PROB ~ ., data = df_tr_bor, method = "lm", trControl=train_control,metric='RMSE')
set.seed(123); rf = train(DEFAULT_PROB ~ ., data = df_tr_bor, method='ranger',
trControl=train_control, tuneGrid = expand.grid(mtry=c(2:5),
min.node.size = 5, splitrule = 'variance'), metric = 'RMSE')
glm_fit = glm(DEFAULT_PROB ~ ., data = df_tr_bor, family = binomial(link = "logit"))
predict(glm_fit)
df_tr_bor$DEFAULT_PROB = (df_tr_bor$DEFAULT_PROB)/100
fit = betareg(DEFAULT_PROB ~ .,data = df_tr_bor)
summary(df_tr_bor$DEFAULT_PROB)
ggplot()
summary(glm_fit)
logit(1.152)
#### OUTLIER DETECTION ####
IsolationForest::IsolationTrees()
IsolationForest::AnomalyScore()