Analyse Random Forest error improvement with additional trees

Hi there,

how do i create something like this in Knime?

Best greetings

1 Like

Hi @Molly123,

Interesting question. I have implemented from scratch the following KNIME workflow to illustrate a possible way to answer your question using KNIME:

The plot below shows that for this particular scholar problem (The Titanic Dataset), a RF made of about 80 trees is performing equally well as RF with more trees:

The workflow can be downloaded from the following KNIME hub link:

Hope it helps.

Best greetings too,


I’ll add a small amendment to @aworker workflow.

KNIME has parameter optimisation nodes which iterate over models to find the best combination of parameters. This can be set up using a Brute Force parameter optimisation to iterate over the number of tree (could also iterate over other parameters as well) to produce the output you seek. The additional benefit is that the final loop node will output the ‘best’ parameter for the model as well as all the other model calculations performed. Also, if all you are looking for is the best parameter combination it can use optimisation algorithms to sample the parameter space and determine the best parameters without needing to do an exhaustive search.

TitanicParameterOptimisation.knwf (65.2 KB)

LinkedIn | Medium | GitHub


This topic was automatically closed 7 days after the last reply. New replies are no longer allowed.