R：随机森林调优超参数的有效方法

Question

I have the following random forest (regression) model with the default parameters

set.seed(42)

# Define train control
trControl <- trainControl(method = "cv",
    number = 10,
    search = "grid")

# Random Forest (regression) model
rf_reg <- train(Price.Gas~., data=data_train,
                    method = "rf",
                    metric = "RMSE",
                    trControl = trControl)

This is the output plot of the true values (black) and the predicted values(red)

I'd like the model to perform better by changing its tunning parameters (eg ntree, maxnodes, search, etc).

I don't think changing one by one is the most efficient way of doing this.

How could I efficiently test the parameters in R to obtain a better random forest (ie one that predicts the data well)?

Answer 1

You will to perform some sort of hyperparameter search (grid or random) where you list all values you want to test (or sequences) and then compute all of them to obtain the best configuration. This links explains the possible aproaches with caret: https://rpubs.com/phamdinhkhanh/389752