Bagged trees

Machine Learning with Tree-Based Models in R

Sandro Raabe

Data Scientist

Many heads are better than one

the wisdom of the crowd

Bootstrap & aggregation

Bagging = short for Bootstrap Aggregation

1. Bootstrapping

Sampling with replacement $\rightarrow$ many modified training sets

2. Aggregation

Predictions of different models are aggregated for final prediction:
- Average (in regression)
- Majority vote (in classification)

Step 1: Bootstrap and train

bootstrapping scheme

Step 2: Aggregate

aggregate results

Coding: Specify the bagged trees

library(baguette)
spec_bagged <- bag_tree() %>%

    set_mode("classification") %>%

    set_engine("rpart", times = 100)

Bagged Decision Tree Model Specification (classification)

Main Arguments:
  cost_complexity = 0
  min_n = 2

Engine-Specific Arguments:
  times = 100

Computational engine: rpart

Train all trees

model_bagged <- fit(spec_bagged, formula = still_customer ~ ., data = customers_train)

parsnip model object

Fit time:  23.9s

Bagged CART (classification with 100 members)

Variable importance scores include:

# A tibble: 19 x 4
   term                     value std.error  used
   <chr>                    <dbl>     <dbl> <int>
 1 total_trans_ct           876.      3.93    100
 2 total_trans_amt          800.      4.54    100
 3 total_revolving_bal      491.      3.67    100

Let's bootstrap!

Machine Learning with Tree-Based Models in R