No Cards or Credit History – Not Ideal
- Having NO Credit history – means No earlier performance reports.
- Having credit history and having managed Credit is BEST.
- Shows can be “Trusted” to pay back what you borrow.
Critical scoring factor - Payment History
- How much and how many times – timely payback.
- Missed payment – BAD Score , Delayed payment – BAD Score again.
- Delayed more than 30 Days – BADLY Scarred.
Utilization or Credit usage Ratio – Lower the better
- How much available credit you use.
- Credit cards have a combined credit limit of 5,00,000 INR.
- Combined outstanding balance of 10,000 INR , thus utilization ratio 2%.
Returning Cards – “Unknown” Bad Karma
- More cards you have – “Lower” the Credit Usage ratio
- Example above – for 2 LACS Combined Limit , 10K is 4%.
- Don’t return those cards.
Some more - “Bad Karma”
- Cash advances – higher return rates and lower Credit Scores.
- Pace out Credit requests – don’t go Multi-Loan Hunting at once.
- Pre-approved instrument is OK – your Credit Seeking is NOT.
### --------------- Technical Stuff Begins ----------------------###
The Logistic Regression Model - I have shared R Code for Logistic regression - whereas Logistic regression has the classic , Classification advantage , where the "Credit Seeker" can be marked as "Good" or "Bad" bet for the bank for a Default on the credit extended - the Logistic Regression is a "Point in Time " algorithm , its a Point Estimate and does not have the capacity to look at an individual portfolio over a period of time . The Logistic Regression model falls in the category of - regression models calledgeneralized linear models.
Initial assessment of the Data - treating missing values . Replacing Outlierswit imputed Values is an inherent initial step - the outliers need not be replaced with the "Mean" or any other point measure of central tendency - the reasons for which have been amply sighted here below in theStats.Statexchange.com post -- http://stats.stackexchange.com/questions/78063/replacing-outliers-with-mean/78065
We would like to look at the Credit Score or Retail Credit Risk of the individual from the standpoint of a time window . Logistic regression does not afford such a model - we may thus look at various other options or a combination of these as listed below -
Stochastic Gradient Descent - Cant be used as Retail Credit Scoring is considered a non-linear classification problem. For the same reason we rule out - Linear Support Vector Machines.
Feature reduction using - Principal Component Analysis - PCA , shall surely be an initial step for any analysis being done. Linear Discriminant Analysis - fisher discriminant analysis are surely to be explored.
Generalized Linear Models with R with special focus on the Logistic Regression model in our case for Credit Scoring - http://data.princeton.edu/R/glms.html
Classification and Regression Trees -- are the Second Algorithm that we would look at after the Initial GLM - a detail about the usage specifics of CART can be seen here -http://www.stat.cmu.edu/~cshalizi/350/lectures/22/lecture-22.pdf
As stated by Daniel Larose in -DATA MINING THE WEB - ZDRAVKO MARKOV AND DANIEL T. LAROSE -- Quote Unlike CART , J. Ross Quinlan’s algorithm C4.5 or J48 is Not restricted to Binary Splits . Whereas CART always produces a binary tree, C4.5 produces a tree of more variable shape. For categorical attributes, C4.5 by default produces a separate branch for each value of the categorical attribute. This may result in more“bushiness” than desired, since some values may have low frequency or maynaturally be associated with other values. UnQuote
Also CART with R can be explored here- http://statweb.stanford.edu/~lpekelis/talks/13_datafest_cart_talk.pdf
Its to be noted that J48 algorithm as used within RWeka and also RATTLE requires that the dataset have all variables as Factors , this if the case the model is then highly over-fit [ i got a very low Error ] i have excluded the R code for this but the same was experimented by me and in case anyone is interested i can share that code and discuss this further.
The Polynomial Kernel based Support Vector Machine - As recommended in most SVM literature - for unbalanced data sets, where one class dominates the other, as is the case with the German Credit Dataset where the Number of Non Defaults or Good cases is much more than the Defaults or Bad Cases - the total Miss-classification Error / error (= sum of Good rejected and Bad accepted divided through the total number of credit clients) is not an appropriate measure. Having said this as seen in the R code and the video shared by me i have duly calculated the Miss-classification Error.
We initially use the SVM with a Radial Basis Function (RBF Kernel )which is the default. Next to better the results of the classification - we use a grid search over all parameters. As recommend we use the tune.svm()function within
library(e1071).
library(e1071).
Upon using grid search over all parameters the best performing SVM can be sought within R .
The code and screen captures seen below are the initial commit this is still Work In Progress ...the search for the best suited algorithm for task at hand continues ...
Partial SVM Code --
gc$default <-factor(gc$default) ;str(gc)
set.seed(123)
train <- sample(1:1000,700)
gc_train <- gc[train,] ; gc_test <- gc[-train,]
str(gc_train)
#
system.time(gc.tune.svm<-tune(svm,default~.,kernal="linear",data=gc_train,ranges=list(gamma=2^(-8:0),cost=10^(-2:4)),scale=T, tunecontrol=tune.control(best.model=TRUE,performances=TRUE, sampling='cross',cross=10)))
summary(gc.tune.svm)
#
#
plot(gc.tune.svm, transform.x = log2, transform.y = log2,main="Contour SVM - German Credit Data",color.palette =terrain.colors)
plot(gc.tune.svm, type = "contour", main="Contour SVM - German Credit Data")
#
print(gc.tune.svm)
#
set.seed(123)
model_gc<-svm(default~.,data=gc_train)
pred<-predict(model_gc,gc_test)
table(pred,gc_test$default)
#
# We want to now see the % of Misclassification by
# creating a Confusion Matrix ..
set.seed(123)
train <- sample(1:1000,700)
gc_train <- gc[train,] ; gc_test <- gc[-train,]
str(gc_train)
#
system.time(gc.tune.svm<-tune(svm,default~.,kernal="linear",data=gc_train,ranges=list(gamma=2^(-8:0),cost=10^(-2:4)),scale=T, tunecontrol=tune.control(best.model=TRUE,performances=TRUE, sampling='cross',cross=10)))
summary(gc.tune.svm)
#
#
plot(gc.tune.svm, transform.x = log2, transform.y = log2,main="Contour SVM - German Credit Data",color.palette =terrain.colors)
plot(gc.tune.svm, type = "contour", main="Contour SVM - German Credit Data")
#
print(gc.tune.svm)
#
set.seed(123)
model_gc<-svm(default~.,data=gc_train)
pred<-predict(model_gc,gc_test)
table(pred,gc_test$default)
#
# We want to now see the % of Misclassification by
# creating a Confusion Matrix ..
# Another approach for the SVM - using the SVM Major package
# GLM - German Credit Data
library(Deducer)## Loading required package: ggplot2
## Loading required package: JGR
## Loading required package: rJava
## Loading required package: JavaGD
## Loading required package: iplots
##
## Please type JGR() to launch console. Platform specific launchers (.exe and .app) can also be obtained at http://www.rforge.net/JGR/files/.
##
##
## Loading required package: car
## Loading required package: MASS
##
##
## Note Non-JGR console detected:
## Deducer is best used from within JGR (http://jgr.markushelbig.org/).
## To Bring up GUI dialogs, type deducer().
##
##
## Attaching package: 'Deducer'
##
## The following object is masked from 'package:stats':
##
## summary.lmlibrary(ggplot2)
#
gcglm<- read.csv("C:/STAT/_Own_R/Credit/Credit-2/gc_names.csv");str(gcglm)## 'data.frame': 1000 obs. of 21 variables:
## $ check_Acc_Status: Factor w/ 4 levels "A11","A12","A13",..: 1 2 4 1 1 4 4 2 4 2 ...
## $ Duration_Months : int 6 48 12 42 24 36 24 36 12 30 ...
## $ Credit_history : Factor w/ 5 levels "A30","A31","A32",..: 5 3 5 3 4 3 3 3 3 5 ...
## $ Credit_purpose : Factor w/ 10 levels "A40","A41","A410",..: 5 5 8 4 1 8 4 2 5 1 ...
## $ amount : int 1169 5951 2096 7882 4870 9055 2835 6948 3059 5234 ...
## $ savings : Factor w/ 5 levels "A61","A62","A63",..: 5 1 1 1 1 5 3 1 4 1 ...
## $ employ.since : Factor w/ 5 levels "A71","A72","A73",..: 5 3 4 4 3 3 5 3 4 1 ...
## $ installment.rate: int 4 2 2 2 3 2 3 2 2 4 ...
## $ status.sex : Factor w/ 4 levels "A91","A92","A93",..: 3 2 3 3 3 3 3 3 1 4 ...
## $ cosigners : Factor w/ 3 levels "A101","A102",..: 1 1 1 3 1 1 1 1 1 1 ...
## $ residence.since : int 4 2 3 4 4 4 4 2 4 2 ...
## $ collateral : Factor w/ 4 levels "A121","A122",..: 1 1 1 2 4 4 2 3 1 3 ...
## $ age : int 67 22 49 45 53 35 53 35 61 28 ...
## $ otherplans : Factor w/ 3 levels "A141","A142",..: 3 3 3 3 3 3 3 3 3 3 ...
## $ housing : Factor w/ 3 levels "A151","A152",..: 2 2 2 3 3 3 2 1 2 2 ...
## $ existing.credits: int 2 1 1 1 2 1 1 1 1 2 ...
## $ job : Factor w/ 4 levels "A171","A172",..: 3 3 2 3 3 2 3 4 2 4 ...
## $ no.dependents : int 1 1 2 2 2 2 1 1 1 1 ...
## $ telephone : Factor w/ 2 levels "A191","A192": 2 1 1 1 1 2 1 2 1 1 ...
## $ foreign : Factor w/ 2 levels "A201","A202": 1 1 1 1 1 1 1 1 1 1 ...
## $ default : int 1 2 1 1 2 1 1 1 1 2 ...View(gcglm)
gcglm$default <-factor(gcglm$default) ;str(gcglm)## 'data.frame': 1000 obs. of 21 variables:
## $ check_Acc_Status: Factor w/ 4 levels "A11","A12","A13",..: 1 2 4 1 1 4 4 2 4 2 ...
## $ Duration_Months : int 6 48 12 42 24 36 24 36 12 30 ...
## $ Credit_history : Factor w/ 5 levels "A30","A31","A32",..: 5 3 5 3 4 3 3 3 3 5 ...
## $ Credit_purpose : Factor w/ 10 levels "A40","A41","A410",..: 5 5 8 4 1 8 4 2 5 1 ...
## $ amount : int 1169 5951 2096 7882 4870 9055 2835 6948 3059 5234 ...
## $ savings : Factor w/ 5 levels "A61","A62","A63",..: 5 1 1 1 1 5 3 1 4 1 ...
## $ employ.since : Factor w/ 5 levels "A71","A72","A73",..: 5 3 4 4 3 3 5 3 4 1 ...
## $ installment.rate: int 4 2 2 2 3 2 3 2 2 4 ...
## $ status.sex : Factor w/ 4 levels "A91","A92","A93",..: 3 2 3 3 3 3 3 3 1 4 ...
## $ cosigners : Factor w/ 3 levels "A101","A102",..: 1 1 1 3 1 1 1 1 1 1 ...
## $ residence.since : int 4 2 3 4 4 4 4 2 4 2 ...
## $ collateral : Factor w/ 4 levels "A121","A122",..: 1 1 1 2 4 4 2 3 1 3 ...
## $ age : int 67 22 49 45 53 35 53 35 61 28 ...
## $ otherplans : Factor w/ 3 levels "A141","A142",..: 3 3 3 3 3 3 3 3 3 3 ...
## $ housing : Factor w/ 3 levels "A151","A152",..: 2 2 2 3 3 3 2 1 2 2 ...
## $ existing.credits: int 2 1 1 1 2 1 1 1 1 2 ...
## $ job : Factor w/ 4 levels "A171","A172",..: 3 3 2 3 3 2 3 4 2 4 ...
## $ no.dependents : int 1 1 2 2 2 2 1 1 1 1 ...
## $ telephone : Factor w/ 2 levels "A191","A192": 2 1 1 1 1 2 1 2 1 1 ...
## $ foreign : Factor w/ 2 levels "A201","A202": 1 1 1 1 1 1 1 1 1 1 ...
## $ default : Factor w/ 2 levels "1","2": 1 2 1 1 2 1 1 1 1 2 ...## Create Design.Matrix or MODEl.Matrix - factor variables, turned to indicator variables
## first column of ones is omitted
set.seed(123)
Xgcglm <- model.matrix(default~.,data=gcglm)[,-1] # Excluded the last "default" variable.
str(Xgcglm) ; Xgcglm[1:10,] ## num [1:1000, 1:48] 0 1 0 0 0 0 0 1 0 1 ...
## - attr(*, "dimnames")=List of 2
## ..$ : chr [1:1000] "1" "2" "3" "4" ...
## ..$ : chr [1:48] "check_Acc_StatusA12" "check_Acc_StatusA13" "check_Acc_StatusA14" "Duration_Months" ...## check_Acc_StatusA12 check_Acc_StatusA13 check_Acc_StatusA14
## 1 0 0 0
## 2 1 0 0
## 3 0 0 1
## 4 0 0 0
## 5 0 0 0
## 6 0 0 1
## 7 0 0 1
## 8 1 0 0
## 9 0 0 1
## 10 1 0 0
## Duration_Months Credit_historyA31 Credit_historyA32 Credit_historyA33
## 1 6 0 0 0
## 2 48 0 1 0
## 3 12 0 0 0
## 4 42 0 1 0
## 5 24 0 0 1
## 6 36 0 1 0
## 7 24 0 1 0
## 8 36 0 1 0
## 9 12 0 1 0
## 10 30 0 0 0
## Credit_historyA34 Credit_purposeA41 Credit_purposeA410
## 1 1 0 0
## 2 0 0 0
## 3 1 0 0
## 4 0 0 0
## 5 0 0 0
## 6 0 0 0
## 7 0 0 0
## 8 0 1 0
## 9 0 0 0
## 10 1 0 0
## Credit_purposeA42 Credit_purposeA43 Credit_purposeA44 Credit_purposeA45
## 1 0 1 0 0
## 2 0 1 0 0
## 3 0 0 0 0
## 4 1 0 0 0
## 5 0 0 0 0
## 6 0 0 0 0
## 7 1 0 0 0
## 8 0 0 0 0
## 9 0 1 0 0
## 10 0 0 0 0
## Credit_purposeA46 Credit_purposeA48 Credit_purposeA49 amount savingsA62
## 1 0 0 0 1169 0
## 2 0 0 0 5951 0
## 3 1 0 0 2096 0
## 4 0 0 0 7882 0
## 5 0 0 0 4870 0
## 6 1 0 0 9055 0
## 7 0 0 0 2835 0
## 8 0 0 0 6948 0
## 9 0 0 0 3059 0
## 10 0 0 0 5234 0
## savingsA63 savingsA64 savingsA65 employ.sinceA72 employ.sinceA73
## 1 0 0 1 0 0
## 2 0 0 0 0 1
## 3 0 0 0 0 0
## 4 0 0 0 0 0
## 5 0 0 0 0 1
## 6 0 0 1 0 1
## 7 1 0 0 0 0
## 8 0 0 0 0 1
## 9 0 1 0 0 0
## 10 0 0 0 0 0
## employ.sinceA74 employ.sinceA75 installment.rate status.sexA92
## 1 0 1 4 0
## 2 0 0 2 1
## 3 1 0 2 0
## 4 1 0 2 0
## 5 0 0 3 0
## 6 0 0 2 0
## 7 0 1 3 0
## 8 0 0 2 0
## 9 1 0 2 0
## 10 0 0 4 0
## status.sexA93 status.sexA94 cosignersA102 cosignersA103 residence.since
## 1 1 0 0 0 4
## 2 0 0 0 0 2
## 3 1 0 0 0 3
## 4 1 0 0 1 4
## 5 1 0 0 0 4
## 6 1 0 0 0 4
## 7 1 0 0 0 4
## 8 1 0 0 0 2
## 9 0 0 0 0 4
## 10 0 1 0 0 2
## collateralA122 collateralA123 collateralA124 age otherplansA142
## 1 0 0 0 67 0
## 2 0 0 0 22 0
## 3 0 0 0 49 0
## 4 1 0 0 45 0
## 5 0 0 1 53 0
## 6 0 0 1 35 0
## 7 1 0 0 53 0
## 8 0 1 0 35 0
## 9 0 0 0 61 0
## 10 0 1 0 28 0
## otherplansA143 housingA152 housingA153 existing.credits jobA172 jobA173
## 1 1 1 0 2 0 1
## 2 1 1 0 1 0 1
## 3 1 1 0 1 1 0
## 4 1 0 1 1 0 1
## 5 1 0 1 2 0 1
## 6 1 0 1 1 1 0
## 7 1 1 0 1 0 1
## 8 1 0 0 1 0 0
## 9 1 1 0 1 1 0
## 10 1 1 0 2 0 0
## jobA174 no.dependents telephoneA192 foreignA202
## 1 0 1 1 0
## 2 0 1 0 0
## 3 0 2 0 0
## 4 0 2 0 0
## 5 0 2 0 0
## 6 0 2 1 0
## 7 0 1 0 0
## 8 1 1 1 0
## 9 0 1 0 0
## 10 1 1 0 0# Print first 10 Rows of MODEl.Matrix.
set.seed(123)
train <- sample(1:1000,900)
# Train set size - 700 ROWS Error - 1.05 ,AIC: 720.27 ,
# Null deviance: 853.51 on 699 degrees of freedom ,
# Residual deviance: 622.27 on 651 degrees of freedom
# Train set size - 800 ROWS Error was - 0.72 ,AIC: 807.01 ,
# Train set size - 900 ROWS Error was - ...AIC: 898.72
# Null deviance: 1094.42 on 899 degrees of freedom
# Residual deviance: 800.72 on 851 degrees of freedom
#
MM_train <- Xgcglm[train,] ; MM_test <- Xgcglm[-train,]
# Training and Testing Data sets from German Credit [MODEl.Matrix]
set.seed(123)
GC_train <- gcglm$default[train] ;GC_test <- gcglm$default[-train]
# Training and Testing Data sets fom German Credit[Data]
# Create Model- GLM, use Train Data from both - GC[MODEl.Matrix] and GC
set.seed(123)
# Families -- binomial(link = "logit")
# quasibinomial(link = "logit")
# quasipoisson(link = "log")
GC_glm_binomial<-glm(default~.,family=binomial,data=data.frame(default=GC_train,MM_train))
#
pdf('GC_glm_binomial.pdf')
rocplot(GC_glm_binomial);# data visualization PDF
dev.off()## png
## 2#
GC_glm_quasibinomial<-glm(default~.,family=quasibinomial,data=data.frame(default=GC_train,MM_train))
pdf('GC_glm_binomial.pdf')
rocplot(GC_glm_quasibinomial);
dev.off()## png
## 2#
# Family == Binomial , link is LOGIT ..
summary(GC_glm_binomial)##
## Call:
## glm(formula = default ~ ., family = binomial, data = data.frame(default = GC_train,
## MM_train))
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.1390 -0.7041 -0.3558 0.7081 2.7150
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) 8.685e-01 1.181e+00 0.736 0.461958
## check_Acc_StatusA12 -3.729e-01 2.305e-01 -1.618 0.105673
## check_Acc_StatusA13 -1.095e+00 3.864e-01 -2.834 0.004600 **
## check_Acc_StatusA14 -1.852e+00 2.516e-01 -7.361 1.82e-13 ***
## Duration_Months 2.920e-02 9.847e-03 2.966 0.003020 **
## Credit_historyA31 -2.374e-01 6.069e-01 -0.391 0.695694
## Credit_historyA32 -8.585e-01 4.675e-01 -1.837 0.066273 .
## Credit_historyA33 -9.974e-01 5.105e-01 -1.954 0.050726 .
## Credit_historyA34 -1.578e+00 4.749e-01 -3.324 0.000887 ***
## Credit_purposeA41 -1.514e+00 3.930e-01 -3.852 0.000117 ***
## Credit_purposeA410 -1.572e+00 8.259e-01 -1.903 0.057036 .
## Credit_purposeA42 -6.360e-01 2.733e-01 -2.327 0.019941 *
## Credit_purposeA43 -6.763e-01 2.617e-01 -2.584 0.009759 **
## Credit_purposeA44 -5.884e-01 9.709e-01 -0.606 0.544447
## Credit_purposeA45 -1.133e-02 5.656e-01 -0.020 0.984025
## Credit_purposeA46 2.137e-01 4.242e-01 0.504 0.614513
## Credit_purposeA48 -2.007e+00 1.258e+00 -1.595 0.110607
## Credit_purposeA49 -8.035e-01 3.611e-01 -2.225 0.026055 *
## amount 9.976e-05 4.673e-05 2.135 0.032783 *
## savingsA62 -4.184e-01 3.101e-01 -1.349 0.177360
## savingsA63 -5.358e-01 4.415e-01 -1.214 0.224873
## savingsA64 -1.152e+00 5.852e-01 -1.969 0.048913 *
## savingsA65 -8.308e-01 2.754e-01 -3.017 0.002555 **
## employ.sinceA72 -6.568e-02 4.626e-01 -0.142 0.887092
## employ.sinceA73 -1.738e-01 4.398e-01 -0.395 0.692690
## employ.sinceA74 -7.611e-01 4.759e-01 -1.599 0.109747
## employ.sinceA75 -1.397e-01 4.413e-01 -0.317 0.751622
## installment.rate 2.976e-01 9.321e-02 3.193 0.001408 **
## status.sexA92 -2.314e-01 4.043e-01 -0.572 0.567086
## status.sexA93 -8.780e-01 3.961e-01 -2.217 0.026655 *
## status.sexA94 -3.234e-01 4.830e-01 -0.670 0.503106
## cosignersA102 3.537e-01 4.263e-01 0.830 0.406697
## cosignersA103 -1.096e+00 4.689e-01 -2.338 0.019385 *
## residence.since -3.934e-02 9.107e-02 -0.432 0.665789
## collateralA122 2.334e-01 2.705e-01 0.863 0.388179
## collateralA123 2.465e-01 2.494e-01 0.989 0.322889
## collateralA124 6.008e-01 4.613e-01 1.302 0.192783
## age -1.299e-02 9.776e-03 -1.329 0.183758
## otherplansA142 -1.888e-01 4.297e-01 -0.439 0.660425
## otherplansA143 -7.551e-01 2.593e-01 -2.912 0.003590 **
## housingA152 -5.307e-01 2.547e-01 -2.083 0.037224 *
## housingA153 -6.189e-01 5.076e-01 -1.219 0.222746
## existing.credits 2.189e-01 1.981e-01 1.105 0.268974
## jobA172 6.618e-01 7.874e-01 0.840 0.400680
## jobA173 6.427e-01 7.643e-01 0.841 0.400399
## jobA174 6.266e-01 7.665e-01 0.817 0.413677
## no.dependents 2.497e-01 2.641e-01 0.945 0.344423
## telephoneA192 -2.256e-01 2.113e-01 -1.067 0.285774
## foreignA202 -1.484e+00 7.113e-01 -2.086 0.036984 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 1094.42 on 899 degrees of freedom
## Residual deviance: 800.72 on 851 degrees of freedom
## AIC: 898.72
##
## Number of Fisher Scoring iterations: 5# Family == Quasi Binomial - attempts to describe additional variance in the data that cannot be explained by a Binomial distribution alone.
summary(GC_glm_quasibinomial) # No AIC Value for family == Quasi witin GLM ..##
## Call:
## glm(formula = default ~ ., family = quasibinomial, data = data.frame(default = GC_train,
## MM_train))
##
## Deviance Residuals:
## Min 1Q Median 3Q Max
## -2.1390 -0.7041 -0.3558 0.7081 2.7150
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 8.685e-01 1.198e+00 0.725 0.468560
## check_Acc_StatusA12 -3.729e-01 2.338e-01 -1.595 0.111105
## check_Acc_StatusA13 -1.095e+00 3.919e-01 -2.793 0.005333 **
## check_Acc_StatusA14 -1.852e+00 2.552e-01 -7.256 8.96e-13 ***
## Duration_Months 2.920e-02 9.990e-03 2.923 0.003554 **
## Credit_historyA31 -2.374e-01 6.157e-01 -0.386 0.699913
## Credit_historyA32 -8.585e-01 4.742e-01 -1.810 0.070584 .
## Credit_historyA33 -9.974e-01 5.179e-01 -1.926 0.054441 .
## Credit_historyA34 -1.578e+00 4.817e-01 -3.277 0.001093 **
## Credit_purposeA41 -1.514e+00 3.986e-01 -3.797 0.000157 ***
## Credit_purposeA410 -1.572e+00 8.379e-01 -1.876 0.061007 .
## Credit_purposeA42 -6.360e-01 2.772e-01 -2.294 0.022016 *
## Credit_purposeA43 -6.763e-01 2.655e-01 -2.547 0.011026 *
## Credit_purposeA44 -5.884e-01 9.849e-01 -0.597 0.550353
## Credit_purposeA45 -1.133e-02 5.738e-01 -0.020 0.984257
## Credit_purposeA46 2.137e-01 4.303e-01 0.496 0.619692
## Credit_purposeA48 -2.007e+00 1.276e+00 -1.573 0.116150
## Credit_purposeA49 -8.035e-01 3.663e-01 -2.194 0.028528 *
## amount 9.976e-05 4.741e-05 2.104 0.035643 *
## savingsA62 -4.184e-01 3.146e-01 -1.330 0.183969
## savingsA63 -5.358e-01 4.479e-01 -1.196 0.231879
## savingsA64 -1.152e+00 5.936e-01 -1.941 0.052551 .
## savingsA65 -8.308e-01 2.794e-01 -2.974 0.003024 **
## employ.sinceA72 -6.568e-02 4.693e-01 -0.140 0.888723
## employ.sinceA73 -1.738e-01 4.461e-01 -0.390 0.696946
## employ.sinceA74 -7.611e-01 4.828e-01 -1.577 0.115270
## employ.sinceA75 -1.397e-01 4.477e-01 -0.312 0.755122
## installment.rate 2.976e-01 9.456e-02 3.148 0.001703 **
## status.sexA92 -2.314e-01 4.102e-01 -0.564 0.572769
## status.sexA93 -8.780e-01 4.018e-01 -2.185 0.029164 *
## status.sexA94 -3.234e-01 4.900e-01 -0.660 0.509386
## cosignersA102 3.537e-01 4.325e-01 0.818 0.413642
## cosignersA103 -1.096e+00 4.757e-01 -2.305 0.021421 *
## residence.since -3.934e-02 9.239e-02 -0.426 0.670374
## collateralA122 2.334e-01 2.744e-01 0.851 0.395212
## collateralA123 2.465e-01 2.530e-01 0.974 0.330106
## collateralA124 6.008e-01 4.680e-01 1.284 0.199547
## age -1.299e-02 9.917e-03 -1.310 0.190434
## otherplansA142 -1.888e-01 4.359e-01 -0.433 0.665075
## otherplansA143 -7.551e-01 2.631e-01 -2.871 0.004198 **
## housingA152 -5.307e-01 2.584e-01 -2.054 0.040317 *
## housingA153 -6.189e-01 5.150e-01 -1.202 0.229739
## existing.credits 2.189e-01 2.009e-01 1.090 0.276162
## jobA172 6.618e-01 7.988e-01 0.828 0.407655
## jobA173 6.427e-01 7.754e-01 0.829 0.407376
## jobA174 6.266e-01 7.776e-01 0.806 0.420585
## no.dependents 2.497e-01 2.679e-01 0.932 0.351604
## telephoneA192 -2.256e-01 2.144e-01 -1.052 0.292991
## foreignA202 -1.484e+00 7.216e-01 -2.056 0.040065 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for quasibinomial family taken to be 1.029107)
##
## Null deviance: 1094.42 on 899 degrees of freedom
## Residual deviance: 800.72 on 851 degrees of freedom
## AIC: NA
##
## Number of Fisher Scoring iterations: 5## Model created now to Predict ...using Test Data.
set.seed(123)
Pred_MM_test <- predict(GC_glm_binomial,newdata=data.frame(MM_test),type="response")
str(Pred_MM_test);head(Pred_MM_test,10)## Named num [1:100] 0.216 0.182 0.117 0.933 0.269 ...
## - attr(*, "names")= chr [1:100] "27" "28" "76" "96" ...## 27 28 76 96 101 104 105
## 0.2162062 0.1821034 0.1168909 0.9327388 0.2688378 0.1291184 0.0291944
## 107 195 219
## 0.5672864 0.3454291 0.6862067str(MM_test)## num [1:100, 1:48] 0 0 0 1 0 1 0 0 1 0 ...
## - attr(*, "dimnames")=List of 2
## ..$ : chr [1:100] "27" "28" "76" "96" ...
## ..$ : chr [1:48] "check_Acc_StatusA12" "check_Acc_StatusA13" "check_Acc_StatusA14" "Duration_Months" ...# Here - Vector=="Pred_MM_test" shown under VALUES within Environment and the Matrix=="MM_test" shown under DATA within Environment.
# Now combine- Vector=="Pred_MM_test" and Matrix=="MM_test" into a DATA.FRAME
Df_GC_glm<-data.frame(MM_test,Pred_MM_test)
# Predict using GLM, use Test Data from both -GC[MODEl.Matrix] and GC
str(Df_GC_glm);head(Df_GC_glm,3)## 'data.frame': 100 obs. of 49 variables:
## $ check_Acc_StatusA12: num 0 0 0 1 0 1 0 0 1 0 ...
## $ check_Acc_StatusA13: num 0 1 0 0 0 0 0 0 0 0 ...
## $ check_Acc_StatusA14: num 1 0 0 0 1 0 1 1 0 0 ...
## $ Duration_Months : num 6 12 12 54 24 9 12 18 45 24 ...
## $ Credit_historyA31 : num 0 1 0 0 0 0 0 1 0 0 ...
## $ Credit_historyA32 : num 0 0 0 0 1 0 1 0 1 1 ...
## $ Credit_historyA33 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ Credit_historyA34 : num 0 0 1 0 0 1 0 0 0 0 ...
## $ Credit_purposeA41 : num 0 0 1 0 0 0 1 0 0 0 ...
## $ Credit_purposeA410 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ Credit_purposeA42 : num 0 0 0 0 0 1 0 0 0 1 ...
## $ Credit_purposeA43 : num 1 1 0 0 0 0 0 0 1 0 ...
## $ Credit_purposeA44 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ Credit_purposeA45 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ Credit_purposeA46 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ Credit_purposeA48 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ Credit_purposeA49 : num 0 0 0 1 0 0 0 0 0 0 ...
## $ amount : num 426 409 1526 15945 1469 ...
## $ savingsA62 : num 0 0 0 0 1 0 0 0 1 0 ...
## $ savingsA63 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ savingsA64 : num 0 1 0 0 0 0 0 0 0 0 ...
## $ savingsA65 : num 0 0 0 0 0 0 1 0 0 0 ...
## $ employ.sinceA72 : num 0 0 0 1 0 0 1 0 0 0 ...
## $ employ.sinceA73 : num 0 1 0 0 0 0 0 0 1 1 ...
## $ employ.sinceA74 : num 0 0 0 0 0 1 0 0 0 0 ...
## $ employ.sinceA75 : num 1 0 1 0 1 0 0 1 0 0 ...
## $ installment.rate : num 4 3 4 3 4 4 2 2 4 2 ...
## $ status.sexA92 : num 0 1 0 0 0 0 0 0 0 0 ...
## $ status.sexA93 : num 0 0 1 1 0 1 0 1 1 0 ...
## $ status.sexA94 : num 1 0 0 0 1 0 1 0 0 0 ...
## $ cosignersA102 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ cosignersA103 : num 0 0 0 0 0 0 0 0 1 0 ...
## $ residence.since : num 4 3 4 4 4 3 4 4 4 2 ...
## $ collateralA122 : num 0 0 0 0 0 0 0 0 1 0 ...
## $ collateralA123 : num 1 0 0 0 0 1 1 0 0 0 ...
## $ collateralA124 : num 0 0 1 1 0 0 0 1 0 0 ...
## $ age : num 39 42 66 58 41 35 26 39 21 24 ...
## $ otherplansA142 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ otherplansA143 : num 1 1 1 1 1 1 1 0 1 1 ...
## $ housingA152 : num 1 0 0 0 0 0 0 1 0 0 ...
## $ housingA153 : num 0 0 1 0 0 0 0 0 0 0 ...
## $ existing.credits : num 1 2 2 1 1 1 1 2 1 1 ...
## $ jobA172 : num 1 0 0 0 1 0 0 0 0 1 ...
## $ jobA173 : num 0 1 0 1 0 1 1 0 1 0 ...
## $ jobA174 : num 0 0 1 0 0 0 0 1 0 0 ...
## $ no.dependents : num 1 1 1 1 1 1 1 2 1 1 ...
## $ telephoneA192 : num 0 0 0 1 0 1 1 1 0 0 ...
## $ foreignA202 : num 0 0 0 0 0 0 0 0 0 0 ...
## $ Pred_MM_test : num 0.216 0.182 0.117 0.933 0.269 ...## check_Acc_StatusA12 check_Acc_StatusA13 check_Acc_StatusA14
## 27 0 0 1
## 28 0 1 0
## 76 0 0 0
## Duration_Months Credit_historyA31 Credit_historyA32 Credit_historyA33
## 27 6 0 0 0
## 28 12 1 0 0
## 76 12 0 0 0
## Credit_historyA34 Credit_purposeA41 Credit_purposeA410
## 27 0 0 0
## 28 0 0 0
## 76 1 1 0
## Credit_purposeA42 Credit_purposeA43 Credit_purposeA44 Credit_purposeA45
## 27 0 1 0 0
## 28 0 1 0 0
## 76 0 0 0 0
## Credit_purposeA46 Credit_purposeA48 Credit_purposeA49 amount savingsA62
## 27 0 0 0 426 0
## 28 0 0 0 409 0
## 76 0 0 0 1526 0
## savingsA63 savingsA64 savingsA65 employ.sinceA72 employ.sinceA73
## 27 0 0 0 0 0
## 28 0 1 0 0 1
## 76 0 0 0 0 0
## employ.sinceA74 employ.sinceA75 installment.rate status.sexA92
## 27 0 1 4 0
## 28 0 0 3 1
## 76 0 1 4 0
## status.sexA93 status.sexA94 cosignersA102 cosignersA103 residence.since
## 27 0 1 0 0 4
## 28 0 0 0 0 3
## 76 1 0 0 0 4
## collateralA122 collateralA123 collateralA124 age otherplansA142
## 27 0 1 0 39 0
## 28 0 0 0 42 0
## 76 0 0 1 66 0
## otherplansA143 housingA152 housingA153 existing.credits jobA172 jobA173
## 27 1 1 0 1 1 0
## 28 1 0 0 2 0 1
## 76 1 0 1 2 0 0
## jobA174 no.dependents telephoneA192 foreignA202 Pred_MM_test
## 27 0 1 0 0 0.2162062
## 28 0 1 0 0 0.1821034
## 76 1 1 0 0 0.1168909## We see in Df Print output - "default" has "probab" and not the earlier 1 OR 2
#
# As these are Randomly Sampled Observations from German Credit data we get Random Row Numbers
# Also the "Pred_MM_test" is the Probability - for example - ....
#
## Mis-classification rates - "GOODS- Will Pay Back" rated as "BADS- Will Default"...
## We use probability cutoff 1/6 or 16.66% , thus we code == Pred_fac<-floor(Pred_MM_test+(5/6))
# if we chose probability cutoff 1/4 or 25.00% ,we code == Pred_fac<-floor(Pred_MM_test+(3/4)).
#
set.seed(123)
Pred_fac<-floor(Pred_MM_test+(5/6))
Pred_fac[1:10]## 27 28 76 96 101 104 105 107 195 219
## 1 1 0 1 1 0 0 1 1 1#
t<-table(GC_test,Pred_fac)# Within GC_test , the 1 is a DEFAULTER - "0" and the 2 NOT DEFAULTER - "1"
t## Pred_fac
## GC_test 0 1
## 1 33 34
## 2 5 28# We want to now see the % of Misclassification by
# creating a Confusion Matrix ...
# As seen -
TN<-33
FN<-5
FP<-34
TP<-28
n_length<-length(Pred_MM_test)
#
# Percentage of Misclassification = (FP+FN)/n_length
Mis.Class<-(FP+FN)/n_length
Mis.Class## [1] 0.39## [1] 0.36 for Pred_MM_test and train <- sample(1:1000,900)
# Percentage of Misclassification = 36%
#
# Sensitivity of Model = TP/(TP+FN)
Sentivity<-TP/(TP+FN)
Sentivity## [1] 0.8484848## [1] 0.8548387 for Pred_MM_test and train <- sample(1:1000,900)
## # Sensitivity of Model = 85.48%
#
# Specificity of Model = TN/(TN+FP)
Specificity<-TN/(TN+FP)
Specificity## [1] 0.4925373## [1] 0.5434783 for Pred_MM_test and train <- sample(1:1000,900)
## # Specificity of Model = 54.34%
#
library(caret)## Loading required package: latticelibrary(ipred)
library(plyr)
library(rpart)
gcBAG<- read.csv("C:/STAT/_Own_R/Credit/Credit-2/gc_names.csv");str(gcBAG)## 'data.frame': 1000 obs. of 21 variables:
## $ check_Acc_Status: Factor w/ 4 levels "A11","A12","A13",..: 1 2 4 1 1 4 4 2 4 2 ...
## $ Duration_Months : int 6 48 12 42 24 36 24 36 12 30 ...
## $ Credit_history : Factor w/ 5 levels "A30","A31","A32",..: 5 3 5 3 4 3 3 3 3 5 ...
## $ Credit_purpose : Factor w/ 10 levels "A40","A41","A410",..: 5 5 8 4 1 8 4 2 5 1 ...
## $ amount : int 1169 5951 2096 7882 4870 9055 2835 6948 3059 5234 ...
## $ savings : Factor w/ 5 levels "A61","A62","A63",..: 5 1 1 1 1 5 3 1 4 1 ...
## $ employ.since : Factor w/ 5 levels "A71","A72","A73",..: 5 3 4 4 3 3 5 3 4 1 ...
## $ installment.rate: int 4 2 2 2 3 2 3 2 2 4 ...
## $ status.sex : Factor w/ 4 levels "A91","A92","A93",..: 3 2 3 3 3 3 3 3 1 4 ...
## $ cosigners : Factor w/ 3 levels "A101","A102",..: 1 1 1 3 1 1 1 1 1 1 ...
## $ residence.since : int 4 2 3 4 4 4 4 2 4 2 ...
## $ collateral : Factor w/ 4 levels "A121","A122",..: 1 1 1 2 4 4 2 3 1 3 ...
## $ age : int 67 22 49 45 53 35 53 35 61 28 ...
## $ otherplans : Factor w/ 3 levels "A141","A142",..: 3 3 3 3 3 3 3 3 3 3 ...
## $ housing : Factor w/ 3 levels "A151","A152",..: 2 2 2 3 3 3 2 1 2 2 ...
## $ existing.credits: int 2 1 1 1 2 1 1 1 1 2 ...
## $ job : Factor w/ 4 levels "A171","A172",..: 3 3 2 3 3 2 3 4 2 4 ...
## $ no.dependents : int 1 1 2 2 2 2 1 1 1 1 ...
## $ telephone : Factor w/ 2 levels "A191","A192": 2 1 1 1 1 2 1 2 1 1 ...
## $ foreign : Factor w/ 2 levels "A201","A202": 1 1 1 1 1 1 1 1 1 1 ...
## $ default : int 1 2 1 1 2 1 1 1 1 2 ...gcBAG$default <- factor(gcBAG$default) ;str(gcBAG)## 'data.frame': 1000 obs. of 21 variables:
## $ check_Acc_Status: Factor w/ 4 levels "A11","A12","A13",..: 1 2 4 1 1 4 4 2 4 2 ...
## $ Duration_Months : int 6 48 12 42 24 36 24 36 12 30 ...
## $ Credit_history : Factor w/ 5 levels "A30","A31","A32",..: 5 3 5 3 4 3 3 3 3 5 ...
## $ Credit_purpose : Factor w/ 10 levels "A40","A41","A410",..: 5 5 8 4 1 8 4 2 5 1 ...
## $ amount : int 1169 5951 2096 7882 4870 9055 2835 6948 3059 5234 ...
## $ savings : Factor w/ 5 levels "A61","A62","A63",..: 5 1 1 1 1 5 3 1 4 1 ...
## $ employ.since : Factor w/ 5 levels "A71","A72","A73",..: 5 3 4 4 3 3 5 3 4 1 ...
## $ installment.rate: int 4 2 2 2 3 2 3 2 2 4 ...
## $ status.sex : Factor w/ 4 levels "A91","A92","A93",..: 3 2 3 3 3 3 3 3 1 4 ...
## $ cosigners : Factor w/ 3 levels "A101","A102",..: 1 1 1 3 1 1 1 1 1 1 ...
## $ residence.since : int 4 2 3 4 4 4 4 2 4 2 ...
## $ collateral : Factor w/ 4 levels "A121","A122",..: 1 1 1 2 4 4 2 3 1 3 ...
## $ age : int 67 22 49 45 53 35 53 35 61 28 ...
## $ otherplans : Factor w/ 3 levels "A141","A142",..: 3 3 3 3 3 3 3 3 3 3 ...
## $ housing : Factor w/ 3 levels "A151","A152",..: 2 2 2 3 3 3 2 1 2 2 ...
## $ existing.credits: int 2 1 1 1 2 1 1 1 1 2 ...
## $ job : Factor w/ 4 levels "A171","A172",..: 3 3 2 3 3 2 3 4 2 4 ...
## $ no.dependents : int 1 1 2 2 2 2 1 1 1 1 ...
## $ telephone : Factor w/ 2 levels "A191","A192": 2 1 1 1 1 2 1 2 1 1 ...
## $ foreign : Factor w/ 2 levels "A201","A202": 1 1 1 1 1 1 1 1 1 1 ...
## $ default : Factor w/ 2 levels "1","2": 1 2 1 1 2 1 1 1 1 2 ...inTrain <- createDataPartition(y=gcBAG$default,p=0.7, list=FALSE)
trn <- gcBAG[inTrain,]
tst <- gcBAG[-inTrain,]
# dim(trn); dim(tst) # Optional
# str(trn);str(tst) # Optional
mFit <- train(default~ .,method="rpart",data=trn)
print(mFit$finalModel)## n= 700
##
## node), split, n, loss, yval, (yprob)
## * denotes terminal node
##
## 1) root 700 210 1 (0.7000000 0.3000000)
## 2) check_Acc_StatusA14>=0.5 275 29 1 (0.8945455 0.1054545) *
## 3) check_Acc_StatusA14< 0.5 425 181 1 (0.5741176 0.4258824)
## 6) Duration_Months< 22.5 243 79 1 (0.6748971 0.3251029) *
## 7) Duration_Months>=22.5 182 80 2 (0.4395604 0.5604396)
## 14) savingsA65>=0.5 26 7 1 (0.7307692 0.2692308) *
## 15) savingsA65< 0.5 156 61 2 (0.3910256 0.6089744) *#
# OK --- library(rattle)
# OK -- fancyRpartPlot(mFit$finalModel)
# GC_bag<-train(default~.,method="treebag",data =gcBAG)
# # str(GC_bag) - DONT ...
# print(GC_bag)
#
# Another Option for Creating Train and Test ...
# library(caret)
# inTrain <- createDataPartition(y=credit$default,p=0.7, list=FALSE)
# trn <- credit[inTrain,]
# tst <- credit[-inTrain,]
# dim(trn); dim(tst)
# str(trn)
#
# Further Reads #
# Quasi Binomial - http://stats.stackexchange.com/questions/91724/what-is-quasibinomial
# https://stat.ethz.ch/R-manual/R-devel/library/stats/html/family.html
# No AIC for Quasi Likelihood or Quasi Binomial
# Akaike's An Information Criterion - https://stat.ethz.ch/R-manual/R-devel/library/stats/html/glm.html
# CRAN Resource - Quasi AIC -- https://cran.r-project.org/web/packages/bbmle/vignettes/quasi.pdf
# SO - http://stackoverflow.com/questions/17045915/using-rocr-package-difficulties
#
# Ignore Code below here .....
# nnn<-1/6
# nnn
# .83333+.16666
#
# old_data <- read.csv("C:/STAT/_Own_R/Credit/Credit-2/d.csv")
# str(old_data)
# qplot(F1.R,F2.R,colour=d,data=trn)
No comments:
Post a Comment