[3K3-J-2] Machine learning: analysis and validations of models

[3K3-J-2-02] On the trade-off between the number of nodes and the number of trees in Random Forest

〇So Kumano¹, Tatsuya Akutsu¹ (1. Kyoto University)

Keywords:Random Forest, expressibility, efficiency

Expressibility of machine learning models has been extensively studied. For example, in a Neural Network, it is
proved that the efficiency concerning the number of nodes is generated from the depth. On the other hand, it is
not clear whether the efficiency exists in Random Forest. Therefore, in this research, we investigate whether the
efficiency exists in Random Forest. We rst show that Random Forest does not have the same kind of efficiency as
Neural Network, and next we show that the efficiency concerning the number of nodes can be generated from the
number of trees.