When splitting a decision tree, the number of samples that the leaves on a branch can possess can become quite low. This can make it hard to be sure that the leaf will do a good job of making good predictions.

This can be dealt with by:

  • Pruning
  • Putting limits on how trees grow by setting a threshold of the number of samples required to split a leaf

Example:

flowchart TD A[Loves popcorn\nSamples: 100] C[Does not love the song\nSamples: 80] D[Loves cats\nSamples: 20] E[Loves the song\nSamples: 19] F[Does not love the song\nSamples: 1] A --> |True| D A --> |False| C D --> |True| E D --> |False| F

References⌗