Now, we will tune RandomForest model. Like SVM, we tune parameter based on 5% downsampling data. The procedure is exactly the same as for SVM model. Below we have reproduced the code for Random Forest model.

set.seed(300)
#down sampling again so than we get more info when stacking
samp = downSample(data_train[-getIndexsOfColumns(data_train, c( "loan_status") )],data_train$loan_status,yname="loan_status")
#choose small data for tuning 
train_index_tuning = createDataPartition(samp$loan_status,p = 0.05,list=FALSE,times=1)
#choose small data for re-train
train_index_training = createDataPartition(samp$loan_status,p = 0.1,list=FALSE,times=1)
rfGrid = expand.grid(
                .mtry = as.integer(seq(2,ncol(samp), (ncol(samp) - 2)/4))
                )
#Install random forest package
library(randomForest)
rfTuned = train(
    samp[train_index_tuning,-getIndexsOfColumns(samp,"loan_status")],
    y = samp[train_index_tuning,"loan_status"],
    method = "rf",
    tuneGrid = rfGrid,
    metric = "ROC",
    trControl = ctrl,
    preProcess = NULL,
    ntree = 100
    )
plot(rfTuned)

> rfTuned
Random Forest 
1268 samples
  70 predictor
   2 classes: 'Default', 'Fully.Paid' 
No pre-processing
Resampling: Cross-Validated (3 fold) 
Summary of sample sizes: 845, 845, 846 
Resampling results across tuning parameters:
  mtry  ROC        Sens       Spec     
   2    0.7028532  0.6909073  0.6199440
  19    0.6832394  0.6451832  0.6088706
  36    0.6706683  0.6231333  0.5820516
  53    0.6748038  0.6263003  0.6026111
  71    0.6751421  0.6609511  0.5962622
ROC was used to select the optimal model using the largest value.
The final value used for the model was mtry = 2.
>

The best parameter is mtry(number of predictors) = 2. Like SVM, we fit 10% of downsampling data with this value.

rf_model = randomForest(loan_status ~ . ,data = samp[train_index_training,],mtry = 2,ntree=400)
predict_loan_status_rf = predict(rf_model,data_test,"prob")
predict_loan_status_rf = as.data.frame(predict_loan_status_rf)$Fully.Paid
rocCurve_rf = roc(response = data_test$loan_status,
               predictor = predict_loan_status_rf)
auc_curve = auc(rocCurve_rf)
plot(rocCurve_rf,legacy.axes = TRUE,print.auc = TRUE,col="red",main="ROC(RandomForest)")

> rocCurve_rf

Call:
roc.default(response = data_test$loan_status, predictor = predict_loan_status_rf)

Data: predict_loan_status_rf in 5358 controls (data_test$loan_status Default) < 12602 cases (data_test$loan_status Fully.Paid).
Area under the curve: 0.705
> 
predict_loan_status_label = ifelse(predict_loan_status_rf<0.5,"Default","Fully.Paid")
c = confusionMatrix(predict_loan_status_label,data_test$loan_status,positive="Fully.Paid")

table_perf[3,] = c("RandomForest",
  round(auc_curve,3),
  as.numeric(round(c$overall["Accuracy"],3)),
  as.numeric(round(c$byClass["Sensitivity"],3)),
  as.numeric(round(c$byClass["Specificity"],3)),
  as.numeric(round(c$overall["Kappa"],3))
  )

The model’s performance is as follow:

> tail(table_perf,1)
         model   auc accuracy sensitivity specificity kappa
3 RandomForest 0.705    0.657       0.666       0.635 0.268
>