Question

In the random forest construction, how many decision tress we need to use to get a good result? How can we use random forest algorithm for regression problem? 

Answer 1

Indeed, the more trees we use, the more accurate the results will be. However, the improvement diminishes somewhat as the number of trees increases, i.e. at some point, the benefit in prediction performance from learning more trees becomes less than the cost in computation time required to learn the additional trees. Random forests, on the other hand, are ensemble methods. Thus, in our case, we observe that a forest of ten trees performs better than a forest of 500 trees in a single experiment on a single test set due to statistical variance.

 

Typical tree counts are 10, 30, or 100, or more appropriately, a random forest should have between 64 and 128 trees. This requires striking a balance between ROC, AUC, and processing time.

 

Random forests can now be used to perform regression tasks as well. Due to the random forest's nonlinear nature, it has an advantage over linear algorithms, making it an excellent choice. However, it is critical to understand that random forests can never extend the data. Random forests can only infer that the data is an average of previously observed values. In a regression problem, the range of predictions that a random forest can make is constrained by the training data's highest and lowest labels.