5. 2016.10.21 SVR, DecisionTree, RandomForest Method

And their Variable Importance using R

• SVR:

  • A prevalent package

    library(e1071)
    fit <- svm(y ~ x1 + x2 + x3 + x4 + x5 + x7 + x8,
              data=sampled_train_data)
    

  • An alternative one which using func Importance() in library rminer

          ​```
          library(rminer)
          fit = fit(y ~ x1 + x2 + x3 + x4 + x5 + x7 + x8, 
                    sampled_train_data, model = 'svm')
          summary(fit) # show results
          im = Importance(fit,sampled_train_data)
          ​```
    

  ​

• Decision Tree:

  library(rpart) fit <- rpart(y ~ x1 + x2 + x3 + x4 + x5 + x7 + x8, method=”anova”, data=sampled_train_data, control=rpart.control(minsplit=100, cp=0) ) summary(fit) # show results fit$variable.importance # importance of each predictor

  • method="anova" for regression model, method="class" for classification model

• Random Forest:

  library(randomForest) fit <- randomForest(y ~ x1 + x2 + x3 + x4 + x5 + x7 + x8, data=sampled_train_data, ntree = 5000) summary(fit) # show results importance(fit) # importance of each predictor