# SVMMAJ - SUPPORT VECTOR MACHINE - AUS CREDIT SCORING
library("SVMMaj", lib.loc="~/R/win-library/3.1")
## Loading required package: kernlab
attach(AusCredit)
str(AusCredit);View(AusCredit)
## List of 2
## $ X:'data.frame': 690 obs. of 14 variables:
## ..$ X1 : logi [1:690] TRUE FALSE FALSE FALSE TRUE FALSE ...
## ..$ X2 : num [1:690] 22.1 22.7 29.6 21.7 20.2 ...
## ..$ X3 : num [1:690] 11.46 7 1.75 11.5 8.17 ...
## ..$ X4 : num [1:690] 2 2 1 1 2 2 2 2 1 2 ...
## ..$ X5 : num [1:690] 4 8 4 5 6 8 3 11 2 4 ...
## ..$ X6 : num [1:690] 4 4 4 3 4 8 4 8 8 8 ...
## ..$ X7 : num [1:690] 1.585 0.165 1.25 0 1.96 ...
## ..$ X8 : logi [1:690] FALSE FALSE FALSE TRUE TRUE TRUE ...
## ..$ X9 : logi [1:690] FALSE FALSE FALSE TRUE TRUE TRUE ...
## ..$ X10: num [1:690] 0 0 0 11 14 2 0 6 0 3 ...
## ..$ X11: logi [1:690] TRUE FALSE TRUE TRUE FALSE FALSE ...
## ..$ X12: num [1:690] 2 2 2 2 2 2 2 2 2 2 ...
## ..$ X13: num [1:690] 100 160 280 0 60 100 60 43 176 100 ...
## ..$ X14: num [1:690] 1213 1 1 1 159 ...
## $ y: Factor w/ 2 levels "Accepted","Rejected": 2 2 2 1 1 1 2 1 2 2 ...
#
# summary(X)
# summary(y)
# detach(AusCredit)
#
?svmmaj
## starting httpd help server ... done
# in SVMMaj - class of each attribute can be either numeric, logical or factor.
# This is Unlike RWeka -
# nnX<-X[1:400,]
# str(nnX) # nnX Data Frame - takes in all variables besides Y , thus X1 to X14 All IN ...
# # X1 : logi , then --- X8 and X9: logi ....All IN till X14
# #
# nny<-y[1:400] # Note - No "," next to 400 as "y" here is a Column Vector ...
# str(nny) # nny Data Frame
#
#
## model training from Aus Credit data , using 400 Obs for Train Set ..
model_SVMMaj <- svmmaj(X[1:400,],y[1:400],hinge='quadratic', lambda=1)
# str(model_SVMMaj)
## model prediction - using balance 290 Obs for Prediction or Test set ..
q4 <- predict(model_SVMMaj,X[-(1:400),],y[-(1:400)],show.plot=TRUE)
q4
## Prediction frequencies:
## Accepted Rejected
## frequency 144 146
##
## Confusion matrix:
## Predicted(yhat)
## Observed (y) Accepted Rejected Total
## Accepted 114 14 128
## Rejected 30 132 162
## Total 144 146 290
##
## Classification Measures:
##
## hit rate 0.848
## misclassification rate 0.152
##
## TP FP Precision
## Accepted 0.891 0.109 0.792
## Rejected 0.815 0.185 0.904
#
## model training-1 , using 500 Obs for Train Set ..
model_SVMMaj1 <- svmmaj(X[1:500,],y[1:500],hinge='quadratic', lambda=1)
# str(model_SVMMaj1)
## model prediction-1 - using balance 190 Obs for Prediction or Test set ..
SVM_1_Plot<- predict(model_SVMMaj1,X[-(1:500),],y[-(1:500)],show.plot=TRUE)
SVM_1_Plot # Here we see correctly Predicted Objects == 88.42% as compared to 84.83% above ...
## Prediction frequencies:
## Accepted Rejected
## frequency 93 97
##
## Confusion matrix:
## Predicted(yhat)
## Observed (y) Accepted Rejected Total
## Accepted 75 4 79
## Rejected 18 93 111
## Total 93 97 190
##
## Classification Measures:
##
## hit rate 0.884
## misclassification rate 0.116
##
## TP FP Precision
## Accepted 0.949 0.0506 0.806
## Rejected 0.838 0.1622 0.959
#
# #yet to be Done ---
# I-spline basis
# model3 <- svmmaj(X,y,weights.obs=weights.obs,spline.knots=3,spline.degree=2)
# plotWeights(model3,plotdim=c(2,4))